scispace - formally typeset
SciSpace - Your AI assistant to discover and understand research papers | Product Hunt

Proceedings ArticleDOI

A Bag of Features Based Approach for Classification of Motile Sperm Cells

01 Jun 2017-pp 104-109

TL;DR: A novel framework based on image processing is proposed to classify sperm cell images affected by noise due to their movement, which allows to achieve classification results with an average accuracy of 90% with the SURF approach compared to 78% with a similar model.

AbstractThe analysis of sperm morphology remains an essential process for diagnosis and treatment of male infertility. In this paper, a novel framework based on image processing is proposed to classify sperm cell images affected by noise due to their movement. This represents a challenge, articularly because the cells are not fixed or stained. The proposed framework is based on Speeded-Up Robust Features (SURF) combined with Bag of Features (BoF) models to quantise features computed by SURF. Support Vector Machines (SVMs) are used to classify the simplified feature vectors, extracted from sperm cell images, into normal, abnormal and noncell categories. The performance of this framework is compared to a similar model where the Histogram of Oriented Gradients (HOG) is used to extract features and SVMs is applied for their classification. The proposed framework allows to achieve classification results with an average accuracy of 90% with the SURF approach compared to 78% with the HOG approach.

Summary (3 min read)

Introduction

  • A novel framework based on image processing is proposed to classify sperm cell images affected by noise due to their movement.
  • This represents a challenge, particularly because the cells are not fixed or stained.
  • Hence, morphology measurements play an important role to select the best sperm cells [10].
  • The rest of this paper is organised as follows: Section II describes foundations of Histogram of Oriented Gradientes (HOG), Speed-Up Robust Features (SURF) and Bag-ofFeatures (BoF) methods and some works related.
  • Experimental results are presented in Section IV and Section V summarises the work.

A. Histogram of Oriented Gradients

  • A Histogram of Oriented Gradients (HOG) was introduced by Dalal et al. [15] for human silhouette detection.
  • Nowadays its use has been extended to other areas such as text and face recognition [16], [17].
  • Orientation of gradients are computed and the histogram of the gradients is calculated.
  • The concatenation of all histograms produces the feature vector of the image.

B. Speeded-Up Robust Features

  • Speeded-Up Robust Features (SURF)—based on the ScaleInvariant Feature Transform (SIFT) [18]—was introduced by Bay et al. [19] and has been shown to be successful in object recognition approaches [20].
  • Mehrotra et al. [21], for instance, use an adaptive SURF descriptor for human iris recognition; Feulner et al. [22] employ SURF for human-body region detection in Computed Tomography (CT) data; and, Han et al. [23] use SURF for traffic sign recognition in colour images.
  • Similarly to SIFT [24], SURF analyses the spatial distribution of gradients.
  • SURF is based on the Hessian matrix and relies on its determinant to select the best response across a range of scales.
  • Hence, it integrates the scale-space theory introduced by Lindeberg [25].

C. Bag of Features

  • Bag of Features (BoF)—originally called Bag of Words (BoW) [26]—was first introduced to natural language processing, text-mining and linguistic methods.
  • Later, BoF is adapted and proposed by Csurka et al. [27] for visual categorisation.
  • A small region surrounding every key point is selected to extract the image features.
  • Finally, quantisation of features yields the histogram representation of the codewords.

A. Template definition

  • Appearance of sperm cells (e.g. area and eccentricity) can change over time due to their movement, the static position of the observer and the representation of a 3D space onto a plane .
  • Note that sperm cells were not stained or altered in the current work.
  • The sperm cell images are categorised into the following classes: Normal.
  • Sperm cell matching the normal morphology criteria and with a sharp head’s edge.
  • Other component in debris or uniform patch.

C. Feature extraction

  • The SURF [19] method is used to detect key points from the images.
  • SURF selects the most representative pixels based in low-level features—using the Hessian matrix.
  • Even though SURF is invariant to rotation, sperm cell samples at different orientations are used to consider the pixellation effect when capturing the images.
  • For every key point, the SURF features are computed using a region of m × n pixels encompassing the key point location yielding to the initial feature vector.

D. Training

  • The number of features is reduced by selecting the minimal number of features n found across the four classes.
  • Thus, the same number of features for each image across all classes can be transferred to the training process of a classifier.
  • In order to do so, the SURF feature vectors are clustered by using K-means.
  • This can be represented by a histogram where the x−axis represents the words and y−axis the number of features mapped to the words .
  • The resulting BoF model is used to train the classifier.

E. Test

  • For each image in the testing dataset, the BoF histogram is computed as follows: Using the BoF model—codebook, a histogram is computed by mapping every SURF feature to a word in the codebook.
  • The distance l2 − norm is used to estimate the closest distance between a feature and the words in the codebook.
  • The resulting histogram is normalised and classified by the SVM model which eventually measures the similarity to the class models obtained in the training process.

A. Experimental Set-up

  • Sperm samples used for this research are obtained from donors at the Academic Unit of Reproductive and Developmental Medicine of the University of Sheffield in the UK.
  • All of donors provided explicit consent to use their samples for research purposes only.
  • Microscopy video sequences are recorded at a rate of 10 frames per second producing 8-bit colour images of 2040 x 1086 pixels in size.
  • Sperm cell samples for both, training and testing, are extracted from frames which were selected by using a random function.
  • A full dataset—images of all classes—is formed by: 235 images labelled as normal; 235 normal-h; 235 abnormal; and, 235 images containing other objects considered as non-cell.

B. Performance evaluation

  • The parameters of the implemented methods are chosen accordingly.
  • The performance obtained with the framework proposed reached an average accuracy of 90%.
  • The average accuracy is defined as the mean between the accuracy of each class, in this case, it is given by the sum of the values along the major diagonal divided by the number of classes, 4.
  • The confusion matrix of the proposed classification method is presented in Table I. Elements along the major diagonal represent the accuracy of the classifier.
  • F −measure = 2× Pre×Rec Pre+Rec (9) The results from this performance measures are summarised in Figure 6.

V. SUMMARY

  • The authors propose a fast and efficient framework for sperm cells classification.
  • The authors approach combines the Bag of Features model where SURF features have been clustered using K-means into a visual words.
  • The analysis comprises the selection of the most representative features that best describe each of the four classes defined in this work for the classification of new instances.
  • The best performance is obtained when the authors define a vocabulary size equal to 2325 keeping 95% of strongest features for the codebook.
  • The proposed framework allows achieving an average classification accuracy of 90% outperforming a similar approach based on HOG for feature extraction which has shown 78% average classification accuracy.

ACKNOWLEDGMENT

  • The authors gratefully acknowledge the financial support of the Mexican Consejo Nacional de Ciencia y Tecnologı́a .
  • The authors also acknowledge the Andrology laboratory of the Academic Unit of Reproductive and Developmental Medicine (University of Sheffield) and especially to Prof. Allan Pacey.

Did you find this useful? Give us your feedback

...read more

Content maybe subject to copyright    Report

This is a repository copy of A Bag of Features Based Approach for Classification of Motile
Sperm Cells.
White Rose Research Online URL for this paper:
http://eprints.whiterose.ac.uk/116761/
Version: Accepted Version
Proceedings Paper:
Davila Garcia, M.L. orcid.org/0000-0001-6259-5781, Paredes Soto, D.A. and Mihaylova,
L.S. (2018) A Bag of Features Based Approach for Classification of Motile Sperm Cells. In:
2017 IEEE International Conference on Internet of Things (iThings) and IEEE Green
Computing and Communications (GreenCom) and IEEE Cyber, Physical and Social
Computing (CPSCom) and IEEE Smart Data (SmartData). 10th IEEE International
Conference on Cyber, Physical and Social Computing (CPSCom-2017), 21-23 Jun 2017,
Exeter, UK. IEEE . ISBN 978-1-5386-3066-2
https://doi.org/10.1109/iThings-GreenCom-CPSCom-SmartData.2017.21
eprints@whiterose.ac.uk
https://eprints.whiterose.ac.uk/
Reuse
Unless indicated otherwise, fulltext items are protected by copyright with all rights reserved. The copyright
exception in section 29 of the Copyright, Designs and Patents Act 1988 allows the making of a single copy
solely for the purpose of non-commercial research or private study within the limits of fair dealing. The
publisher or other rights-holder may allow further reproduction and re-use of this version - refer to the White
Rose Research Online record for this item. Where records identify the publisher as the copyright holder,
users can verify any specific terms of use on the publisher’s website.
Takedown
If you consider content in White Rose Research Online to be in breach of UK law, please notify us by
emailing eprints@whiterose.ac.uk including the URL of the record and the reason for the withdrawal request.

A Bag of Features Based Approach for
Classification of Motile Sperm Cells
Maria Luisa Davila Garcia, Daniel A. Paredes Soto, Lyudmila S. Mihaylova,
Department of Automatic Control and Systems Engineering, The University of Sheffield, UK
{mdavilagarcia1, daniel.paredes-soto, l.s.mihaylova}@sheffield.ac.uk
Abstract—The analysis of sperm morphology remains an
essential process for diagnosis and treatment of male infertility.
In this paper, a novel framework based on image processing
is proposed to classify sperm cell images affected by noise due to
their movement. This represents a challenge, particularly because
the cells are not fixed or stained.
The proposed framework is based on Speeded-Up Robust
Features (SURF) combined with Bag of Features (BoF) models to
quantise features computed by SURF. Support Vector Machines
(SVMs) are used to classify the simplified feature vectors, ex-
tracted from sperm cell images, into normal, abnormal and non-
cell categories. The performance of this framework is compared
to a similar model where the Histogram of Oriented Gradients
(HOG) is used to extract features and SVMs is applied for their
classification.
The proposed framework allows to achieve classification results
with an average accuracy of 90% with the SURF approach
compared to 78% with the HOG approach.
Index Terms—sperm cell; morphology; classification; SURF;
HOG; bag of visual features; feature extraction
I. INTRODUCTION
Medical systems require processing of a large amount of
data. However, this poses a significant challenge to medical
professionals and makes even impossible for a human to anal-
yse this information precisely and reliably. This paper presents
an approach for the automated analysis and classification of
motile sperm cells.
Sperm cell concentration, morphology and motility esti-
mates are, among other parameters, essential for diagnosis and
treatment of male infertility according to the World Health
Organization (WHO) [1]. Nowadays, analysis of sperm cells
remains subjective, imprecise, and highly time-consuming task
[2]. A health professional, usually using a microscope, counts
and evaluates the morphology of cells following guidelines es-
tablished by the WHO [1] and based on their own experience.
Computer-Assisted Semen Analysis (CASA) systems have
been used at andrology laboratories and Assisted Reproductive
Units (ARU) world-wide in the last decades. Nevertheless,
CASA systems represent a high-cost and frequently lack for
adaptive methods able to work under a variety of conditions,
since sperm sample vary significantly.
Over the last years, a wide range of works aim to achieve
accurate and robust morphology classification of sperm cell.
These include not only human sperm cell, but also different
species [3], [4] and other biological particles: virus, bacteria,
stem cells [5] and tumour images [6].
Methods for assisted conception such as Intra-
Cytoplasmatic Sperm Injection (ICSI) and in-vitro fertilisation
(IVF) require a selection of a few sperm cells for the
fertilisation. In ICSI, for instance, an embryologist selects the
best normal-looking sperm cells under high magnification.
It has been found that morphology can be associated to
the sperm cell function including the ability for fertilisation
(based on statistical estimates only) [7], [8], [9]. Hence,
morphology measurements play an important role to select
the best sperm cells [10].
Motile sperm cells can be labelled as normal or abnormal
based on its morphology, by fulfilling the strict WHO refe-
rences [1], and their motility grade. A formal definition for
normal morphology of sperm cells is presented by Menkveld
et al. [11] and it is described as follows. The head is if length
between 3 and 5 µm and the width is between 2 and 3 µm. The
head has to contain a well-defined acrosomal region and the
sperm tail measuring 45 µm in length with a uniform width.
Also, motility of sperm cells can be affected by physical
characteristics of the seminal fluid (e.g. viscosity and pH),
sperm cell variables (e.g. sperm concentration) and the pre-
sence of debris [1]. Those conditions can produce noise and
changes in illumination when images are captured making
automated sperm cell detection difficult. In addition, we need
to consider the noise introduced by imaging devices during
the image formation process in the hardware used.
Staining procedures using compounds have been reported
improving sperm cell segmentation due to the large colour
contrast produced between live sperm cells and the background
in images. Nevertheless, staining of sperm cells have also been
reported to cause morphology alterations, specifically changes
in their head [12].
Appearance of sperm cells in video sequences can change
due to the conditions at every scene captured. Variation of
illumination resulting from the flow of the seminal fluid, for
instance, can produce shades in the background. Focus drift
from the temperature gradient and mechanical vibrations in
the microscope can affect the sharpness of the elements in
the image [13], [14]. Also, the movement of sperm cells can
occur at any orientation in samples where no attracting source
is utilised. Thus, the shape and size of sperm cells can be
visually different when they move orthogonally to the camera
plane.
The framework presented is this paper integrates SURF, Bag
of Features and Support Vector Machine for the classification

of sperm cell in microscopy images based on low-level fea-
tures. The approach proposed is able to analyse images from
unstained and motile sperm cells.
The rest of this paper is organised as follows: Section
II describes foundations of Histogram of Oriented Gradi-
entes (HOG), Speed-Up Robust Features (SURF) and Bag-of-
Features (BoF) methods and some works related. In Section
III, the proposed approach is developed. Experimental results
are presented in Section IV and Section V summarises the
work.
II. RELATED WORKS
Digital image processing methods can be used to detect
and classify objects based on low-level features in microscopy
images. The approach proposed in this paper combines a set
of algorithms to extract and organise features of sperm cell
images to train a classifier. A SVM algorithm is used to
classify features from new image samples into one of the
following classes: normal-, abnormal- and non-sperm cell.
A. Histogram of Oriented Gradients
A Histogram of Oriented Gradients (HOG) was introduced
by Dalal et al. [15] for human silhouette detection. Nowadays
its use has been extended to other areas such as text and face
recognition [16], [17]. The HOG is based on the distribution of
intensities of gradients G
x
and G
y
of a given image. Gradient
estimates at a pixel(i, j) are given by (1) and (2).
G
x(i,j)
= f (i + 1, j) f (i 1, j) , (1)
G
y(i,j)
= f (i, j + 1) f (i, j 1) (2)
where f (i, j) is the intensity value at pixel location (i, j).
Gradients can be used to estimate the local orientation θ and
magnitude H of the gradient:
θ (i, j) = arctan(G
x(i,j)
/G
y(i,j)
), (3)
H (i, j) =
q
G
2
x(i,j)
+ G
2
y(i,j)
(4)
To compute the HOG of a given image, it is divided into
small regions termed cells. Orientation of gradients are com-
puted and the histogram of the gradients is calculated. Gradient
can be performed by filtering the image using a Sobel-based
kernel D
x
= [1 0 1] and D
y
= [1 0 1]
T
. The
concatenation of all histograms produces the feature vector
of the image.
A visualisation of HOG features extracted from a normal
sperm cell using a patch size of 4 × 4 and 8 × 8 pixel units is
shown in Figure 1.
B. Speeded-Up Robust Features
Speeded-Up Robust Features (SURF)—based on the Scale-
Invariant Feature Transform (SIFT) [18]—was introduced by
Bay et al. [19] and has been shown to be successful in object
recognition approaches [20].
Mehrotra et al. [21], for instance, use an adaptive SURF
descriptor for human iris recognition; Feulner et al. [22]
(a)
(b) (c)
Fig. 1. Visualization of HOG features: (a) Original image, (b) HOG using a
patch size of 4 × 4 and (c) using a patch size of 8 × 8.
employ SURF for human-body region detection in Computed
Tomography (CT) data; and, Han et al. [23] use SURF for
traffic sign recognition in colour images.
Similarly to SIFT [24], SURF analyses the spatial distribu-
tion of gradients. In addition, SURF divides the image in sub-
regions that make the method faster and less noise-sensitive
[19]. SURF is based on the Hessian matrix and relies on
its determinant to select the best response across a range of
scales. Hence, it integrates the scale-space theory introduced
by Lindeberg [25]. The Hessian matrix H (x, σ) at a location
with pixel coordinates x = (x, y) and scale σ in an image I
is given by (5).
H (x, σ) =
L
xx
(x, σ) L
xy
(x, σ)
L
xy
(x, σ) L
yy
(x, σ)
(5)
where L
xx
(x, σ) is the second-order partial derivative
2
x
2
g(α) with the image I at point x and can be estimated
by the convolution of I with a second-order derivative of a
Gaussian kernel, also known as Laplacian of Gaussian (LoG);
L
yy
(x, σ) and L
xy
(x, σ) can be estimated similarly. Unlike
SIFT, SURF estimates the second-order Gaussian derivatives
using box filters based on integral images.
C. Bag of Features
Bag of Features (BoF)—originally called Bag of Words
(BoW) [26]—was first introduced to natural language process-
ing, text-mining and linguistic methods. Later, BoF is adapted
and proposed by Csurka et al. [27] for visual categorisation.
BoF has been widely used. Shen et al. [28], for instance, use
BoW for classification of cells in biomedical images. Zhou
[29] and Nanni et al. [30] employ BoF to classify scenes
contained in images.
BoF aims to represent the extracted features of an image
as a histogram of the computed features by quantising the
features. The BoF model for sperm cell images classification
is summarised in the following steps:

1) Selection of features. In this paper, the SURF method
is used to extract the strongest key points representing
an image. A small region (patch) surrounding every key
point is selected to extract the image features.
2) Learning vocabulary—also termed visual codebook. In
this step, the extracted features are divided into groups
(clusters). The clustering process can be carried out by
using the K-means approach, where the centroid of a
cluster represents a visual word of the codebook [31].
3) Feature quantisation. The final step in the BoF model
is the mapping process of every feature (patch) into a
specific codeword by using a distance metric (e.g. Man-
hattan or Euclidean). Finally, quantisation of features
yields the histogram representation of the codewords.
III. THE PROPOSED FRAMEWORK FOR SPERM CELLS
CLASSIFICATION
The proposed method for sperm cell classification is sum-
marised in Figure 2 and each step is described in the following
sections.
Video sequence
Selection of frames
containing sperm-cells and other
elements in the background
Extract image samples for each class
Normal Normal-h Abormal
Non sperm-cell
Feature extraction
Key-point
detection (SURF)
Feature
extraction
(SURF)
Training
Bag-of-Features +
Support-Vector-Machine (SVM)
Selection of the
strongest features
Visual word vocabulary,
clustering with K -means
(Codebook)
. . .
BoF histogram
representation per class
Train classifier (SVM)
Test
BoF histogram
representation
Similarity evaluation
Outcome
{Normal, Normal-h,
Abnormal, Non sperm-cell}
Fig. 2. Proposed framework for unstained sperm cell classification based on
SURF, Bag-of-Features and SVM classifier.
A. Template definition
Appearance of sperm cells (e.g. area and eccentricity) can
change over time due to their movement, the static position of
the observer and the representation of a 3D space onto a plane
(image). In Figure 3, for instance, a set of two normal sperm
cells along a sequence of five consecutive frames is shown.
The variation in size and brightness can be observed as sperm
cells swim. Note that sperm cells were not stained or altered
in the current work. Those visual artefacts can challenge the
methods used to automate the classification of sperm cells.
A
B
t t+1 t+2 t+3 t+4
Fig. 3. Appearance variation of two sperm cells (upper and lower row) along
consecutive video frames t [0, 1, ..., 4].
In Figure 3, a ‘halo’ surrounding sperm’s head can be
observed in some images making head’s edge difficult to
distinguish. Therefore, the variation in the sharpness of sperm-
cell’s head is considered into the definition of a normal sperm
cell in this paper.
The sperm cell images are categorised into the following
classes:
Normal. Sperm cell matching the normal morphology
criteria and with a sharp head’s edge.
Normal-h. Normal sperm cell with a blurred head’s edge.
Abnormal. Sperm cell that does not meet the normal
morphology definition.
Other. Other component in debris or uniform patch.
Figure 4 shows sperm cell examples of the classes described
above.
B. Datasets
A collection of video frames showing sperm cells and other
objects are selected to create two datasets: training and testing.
Image patches containing a single sperm cell are selected
for the different classes defined. The patches are manually
selected to include the possible variates that can be found in
practice (e.g. orientation and morphology of sperm cells and
other objects in the background). The testing dataset is used to
validate and measure the performance of the method proposed.
C. Feature extraction
The SURF [19] method is used to detect key points from
the images. SURF selects the most representative pixels based
in low-level features—using the Hessian matrix. Even though
SURF is invariant to rotation, sperm cell samples at different
orientations are used to consider the pixellation effect when
capturing the images. For every key point, the SURF features

(a) (b) (c)
(d) (e) (f)
(g) (h) (i)
Fig. 4. Sperm cell samples from the four categories defined: (a),(b) normal;
(c) normal-h; (d)-(f) abnormal; (g)-(i) other components in debris.
are computed using a region of m × n pixels encompassing
the key point location yielding to the initial feature vector.
D. Training
1) Selection of the strongest features: The number of fea-
tures is reduced by selecting the minimal number of features
n found across the four classes. The SURF method is used
to choose the n strongest features from each image. Thus, the
same number of features for each image across all classes can
be transferred to the training process of a classifier.
2) Bag-of-Features (BoF): Processing large SURF vectors
can represent a high computational cost. Hence, the BoF
model is employed to reduce the feature representation of the
images. The aim of BoF is creating a codebook—also termed
Visual Word Vocabulary—by transforming feature vectors into
visual words. In order to do so, the SURF feature vectors
are clustered by using K-means. The number of clusters K
defines the codebook size and the centroids of clusters define
the words.
Thus, each feature of the SURF vectors is assigned to
one word. This can be represented by a histogram where
the xaxis represents the words and yaxis the number of
features mapped to the words (see Figure 5). The BoF method
is repeated to produce the histogram of BoF for each of the
four classes defined. The resulting BoF model is used to train
the classifier.
3) Training of a Support Vector Machine (SVM) classifier:
A multi-class Support Vector Machine (SVM) classifier is
trained with the BoF model obtained. SVMs are inherently
binary classifiers. Thus, an approach is set-up to define the
class for a new entry. A ‘one-to-one’ model is used to define
the outcome based on the most voted class by the binary
classifiers. The trained SVM model is used to process BoF
histograms instances from the training dataset.
(a)
100 200 300 400 500
Visual word index
Frequency of occurrence
(b)
Fig. 5. Normalised BoF histogram using SURF features from a non-cell
sample image. (a) Original image, (b) BoF histogram using 500 words.
E. Test
For each image in the testing dataset, the BoF histogram
is computed as follows: Using the BoF model—codebook, a
histogram is computed by mapping every SURF feature to a
word in the codebook. The distance l
2
norm is used to
estimate the closest distance between a feature and the words
in the codebook. The resulting histogram is normalised and
classified by the SVM model which eventually measures the
similarity to the class models obtained in the training process.
IV. EXPERIMENTS AND RESULTS
A. Experimental Set-up
Sperm samples used for this research are obtained from
donors at the Academic Unit of Reproductive and Develop-
mental Medicine (AURDM) of the University of Sheffield in
the UK. All of donors provided explicit consent to use their
samples for research purposes only. The sperm samples were
not prepared or pre-processes (diluted, stained or altered to
re-orientate their direction.)
Microscopy video sequences are recorded at a rate of 10
frames per second producing 8-bit colour images of 2040 x
1086 pixels in size. Sperm cell samples for both, training
and testing, are extracted from frames which were selected
by using a random function. A full dataset—images of all
classes—is formed by: 235 images labelled as normal; 235
normal-h; 235 abnormal; and, 235 images containing other
objects considered as non-cell.
The labelling process is based on the WHO guidelines and
the visual appearance of the sperm cell images. The images
ranged from 60 × 60 to 70 × 85 pixels in size and are
compressed into JPEG (Joint Photographic Experts Group)
format.
In this paper, 235 images are used for each class. This
process aims to make a fear comparison between the number
of features for each class in the training process.
The full dataset is divided into two datasets: 30% of the
images were used for training and 70% for testing.
B. Performance evaluation
The parameters of the implemented methods are chosen
accordingly. The 95% of the strongest SURF features are used
to choose a reduced number of features which are used by the

Citations
More filters

01 Jan 2013
TL;DR: A computational framework that tracks the heads and traces the tails for analyzing sperm motility, one of the most important attributes in semen quality evaluation, and identifies other existing methods based merely on the head trajectories.
Abstract: Sperm quality assessment plays an essential role in human fertility and animal breeding. Manual analysis is time-consuming and subject to intra- and inter-observer variability. To automate the analysis process, as well as to offer a means of statistical analysis that may not be achieved by visual inspection, we present a computational framework that tracks the heads and traces the tails for analyzing sperm motility, one of the most important attributes in semen quality evaluation. Our framework consists of 3 modules: head detection, head tracking, and flagellum tracing. The head detection module detects the sperm heads from the image data, and the detected heads are the inputs to the head tracking module for obtaining the head trajectories. Finally, a flagellum tracing algorithm is proposed to obtain the flagellar beat patterns. Our framework aims at providing both the head trajectories and the flagellar beat patterns for quantitatively assessing sperm motility. This distinguishes our work from other existing methods that analyze sperm motility based merely on the head trajectories. We validate our framework using two confocal microscopy image sequences of ram semen samples that were imaged at two different conditions, at which the sperms behave differently. The results show the effectiveness of our framework.

15 citations


Book ChapterDOI
25 Jun 2019
TL;DR: A ORB-based fixed multi-resolution recognition algorithm that achieves over 95% accuracy at a resolution scale of 0.2 and an approximately 60% faster recognition time than the next best comparable method is proposed.
Abstract: Maintenance and troubleshooting of hardware on a large scale pose a challenge in deploying expert technicians at multiple sites. Augmented Reality-based technology support equips the technicians with the skills they need to solve hardware problems even without expert level training, thereby reducing training time and cost to the vendor. Enabling Augmented Reality for technology support requires the ability to visually recognize the hardware in real time using mobile devices, and train the underlying algorithms at scale. This paper proposes a novel approach to address these issues. Our ORB-based fixed multi-resolution recognition algorithm achieves over 95% accuracy at a resolution scale of 0.2, and an approximately 60% faster recognition time than the next best comparable method. We also demonstrate the real-world applicability of our algorithm through an implementation of an Augmented Reality application.

1 citations


Journal ArticleDOI
Abstract: Recent days, the infertility is affecting one of every ten couples. This makes the negative effect on the quality of a couple’s life, social causes, and psychological problems. Here the sperm morphology analysis helps to diagnose this problem. Here the machine learning approach is used for classification, detection and segmentation process. This also utilizes morphology approach for image representation. In this proposed method, the deep convolutional neural network is used for detecting the abnormality of human male infertility. Here the image morphological process is employed with the enhanced Otsu’s threshold method for segmenting the sperm image, which helps to detect the abnormal region using convolution layer. Here the database is collected from the human sperm image analysis dataset. Initially, the morphological process is applied to reduce the noise from the given set of input image then the segmentation process is performed by using E-Otsu’s threshold method. Two-dimensional Otsu’s thresholding technique reduces the computation complexity and it uses the median filter and for edge reduction approach sobel operator is used, which improves the performance of segmentation. Overall, the proposed research work optimizes three sections that are image representation by morphology approach, image segmentation by Enhanced-Otsu’s thresholding approach, and abnormality detection by Convolutional Neural Network. This method obtains the result of accuracy, detection rate, and computation time. By comparing with the existing method, the proposed method achieves the 98.99% of accuracy result and detects the abnormality effectively with the reduced computation time of 4 min and 15 s. This proposed work is done by using MATLAB with the adaptation of 2018a.

References
More filters

Journal ArticleDOI
TL;DR: This paper presents a method for extracting distinctive invariant features from images that can be used to perform reliable matching between different views of an object or scene and can robustly identify objects among clutter and occlusion while achieving near real-time performance.
Abstract: This paper presents a method for extracting distinctive invariant features from images that can be used to perform reliable matching between different views of an object or scene. The features are invariant to image scale and rotation, and are shown to provide robust matching across a substantial range of affine distortion, change in 3D viewpoint, addition of noise, and change in illumination. The features are highly distinctive, in the sense that a single feature can be correctly matched with high probability against a large database of features from many images. This paper also describes an approach to using these features for object recognition. The recognition proceeds by matching individual features to a database of features from known objects using a fast nearest-neighbor algorithm, followed by a Hough transform to identify clusters belonging to a single object, and finally performing verification through least-squares solution for consistent pose parameters. This approach to recognition can robustly identify objects among clutter and occlusion while achieving near real-time performance.

42,225 citations


Proceedings ArticleDOI
20 Sep 1999
TL;DR: Experimental results show that robust object recognition can be achieved in cluttered partially occluded images with a computation time of under 2 seconds.
Abstract: An object recognition system has been developed that uses a new class of local image features. The features are invariant to image scaling, translation, and rotation, and partially invariant to illumination changes and affine or 3D projection. These features share similar properties with neurons in inferior temporal cortex that are used for object recognition in primate vision. Features are efficiently detected through a staged filtering approach that identifies stable points in scale space. Image keys are created that allow for local geometric deformations by representing blurred image gradients in multiple orientation planes and at multiple scales. The keys are used as input to a nearest neighbor indexing method that identifies candidate object matches. Final verification of each match is achieved by finding a low residual least squares solution for the unknown model parameters. Experimental results show that robust object recognition can be achieved in cluttered partially occluded images with a computation time of under 2 seconds.

15,597 citations


"A Bag of Features Based Approach fo..." refers methods in this paper

  • ...Unlike SIFT, SURF estimates the second-order Gaussian derivatives using box filters based on integral images....

    [...]

  • ...Similarly to SIFT [24], SURF analyses the spatial distribution of gradients....

    [...]

  • ...Speeded-Up Robust Features (SURF)—based on the ScaleInvariant Feature Transform (SIFT) [18]—was introduced by Bay et al. [19] and has been shown to be successful in object recognition approaches [20]....

    [...]

  • ...Speeded-Up Robust Features (SURF)—based on the ScaleInvariant Feature Transform (SIFT) [18]—was introduced by Bay et al....

    [...]


01 Jan 2011
TL;DR: The Scale-Invariant Feature Transform (or SIFT) algorithm is a highly robust method to extract and consequently match distinctive invariant features from images that can then be used to reliably match objects in diering images.
Abstract: The Scale-Invariant Feature Transform (or SIFT) algorithm is a highly robust method to extract and consequently match distinctive invariant features from images. These features can then be used to reliably match objects in diering images. The algorithm was rst proposed by Lowe [12] and further developed to increase performance resulting in the classic paper [13] that served as foundation for SIFT which has played an important role in robotic and machine vision in the past decade.

14,701 citations


"A Bag of Features Based Approach fo..." refers methods in this paper

  • ...Unlike SIFT, SURF estimates the second-order Gaussian derivatives using box filters based on integral images....

    [...]

  • ...Similarly to SIFT [24], SURF analyses the spatial distribution of gradients....

    [...]

  • ...Speeded-Up Robust Features (SURF)—based on the ScaleInvariant Feature Transform (SIFT) [18]—was introduced by Bay et al. [19] and has been shown to be successful in object recognition approaches [20]....

    [...]


Journal ArticleDOI
TL;DR: A novel scale- and rotation-invariant detector and descriptor, coined SURF (Speeded-Up Robust Features), which approximates or even outperforms previously proposed schemes with respect to repeatability, distinctiveness, and robustness, yet can be computed and compared much faster.
Abstract: This article presents a novel scale- and rotation-invariant detector and descriptor, coined SURF (Speeded-Up Robust Features). SURF approximates or even outperforms previously proposed schemes with respect to repeatability, distinctiveness, and robustness, yet can be computed and compared much faster. This is achieved by relying on integral images for image convolutions; by building on the strengths of the leading existing detectors and descriptors (specifically, using a Hessian matrix-based measure for the detector, and a distribution-based descriptor); and by simplifying these methods to the essential. This leads to a combination of novel detection, description, and matching steps. The paper encompasses a detailed description of the detector and descriptor and then explores the effects of the most important parameters. We conclude the article with SURF's application to two challenging, yet converse goals: camera calibration as a special case of image registration, and object recognition. Our experiments underline SURF's usefulness in a broad range of topics in computer vision.

11,276 citations


"A Bag of Features Based Approach fo..." refers background or methods in this paper

  • ...In addition, SURF divides the image in subregions that make the method faster and less noise-sensitive [19]....

    [...]

  • ...[19] and has been shown to be successful in object recognition approaches [20]....

    [...]

  • ...The SURF [19] method is used to detect key points from the images....

    [...]


Proceedings Article
01 Jan 2004
TL;DR: This bag of keypoints method is based on vector quantization of affine invariant descriptors of image patches and shows that it is simple, computationally efficient and intrinsically invariant.
Abstract: We present a novel method for generic visual categorization: the problem of identifying the object content of natural images while generalizing across variations inherent to the object class. This bag of keypoints method is based on vector quantization of affine invariant descriptors of image patches. We propose and compare two alternative implementations using different classifiers: Naive Bayes and SVM. The main advantages of the method are that it is simple, computationally efficient and intrinsically invariant. We present results for simultaneously classifying seven semantic visual categories. These results clearly demonstrate that the method is robust to background clutter and produces good categorization accuracy even without exploiting geometric information.

4,911 citations


Frequently Asked Questions (1)
Q1. What have the authors contributed in "A bag of features based approach for classification of motile sperm cells" ?

In this paper, a novel framework based on image processing is proposed to classify sperm cell images affected by noise due to their movement. The performance of this framework is compared to a similar model where the Histogram of Oriented Gradients ( HOG ) is used to extract features and SVMs is applied for their classification.