Enhanced Local Texture Feature Sets for Face Recognition Under Difficult Lighting Conditions
read more
Citations
A Completed Modeling of Local Binary Pattern Operator for Texture Classification
Enhanced Computer Vision With Microsoft Kinect Sensor: A Review
PCANet: A Simple Deep Learning Baseline for Image Classification?
PCANet: A Simple Deep Learning Baseline for Image Classification?
WLD: A Robust Local Image Descriptor
References
Histograms of oriented gradients for human detection
Nonlinear total variation based noise removal algorithms
Eigenfaces for recognition
Multiresolution gray-scale and rotation invariant texture classification with local binary patterns
Eigenfaces vs. Fisherfaces: recognition using class specific linear projection
Related Papers (5)
A comparative study of texture measures with classification based on featured distributions
Frequently Asked Questions (13)
Q2. What are the main advantages of their method?
The main advantages of their method are simplicity, computational efficiency and robustness to lighting changes and other image quality degradations such as blurring.
Q3. How long does it take to process a face image?
Their (unoptimized Matlab) implementation takes only about 50 ms to process a 120×120 pixel face image on a 2.8 GHz P4, allowing face preprocessing to be performed in real time.
Q4. What is the main reason why the authors have chosen to normalize the image?
Since run time is a critical factor in many practical applications, it is also interesting to consider the computational load of their normalization chain.
Q5. How many pixels can be mapped to a given target?
For a given target the transform can be computed and mapped through w() in a preprocessing step, after which matching to any subsequent image takes O(number of pixels) irrespective of the number of code values.
Q6. What are the main criticisms of the LBP method?
Possible criticisms of this method are that subdividing the face into a regular grid is somewhat arbitrary (cells are not necessarily well aligned with facial features), and that partitioning appearance descriptors into grid cells is likely to cause both aliasing (due to abrupt spatial quantization) and loss of spatial resolution (as position within a grid cell is not coded).
Q7. What is the way to offset the center of the larger filter?
for some datasets it also helps to offset the center of the larger filter by 1–2 pixels relative to the center of the smaller one, so that the final prefilter is effectively the sum of a centered DoG and a low pass spatial derivative.
Q8. What is the effect of the nonlinear function on the LBP feature set?
To reduce their influence on subsequent stages of processing, the authors finally apply a nonlinear function to compress over-large values.
Q9. How many of the 256 8-bit patterns are uniform?
Ojala et al. observed that although only 58 of the 256 8-bit patterns are uniform,nearly 90 percent of all observed image neighbourhoods are uniform.
Q10. What is the local binary pattern operator?
The operator takes a local neighborhood around each pixel, thresholds the pixels of the neighborhood at the value of the central pixel and uses the resulting binary-valued image patch as a local image descriptor.
Q11. How many pixels are used to place the center of the two eyes?
All of the images undergo the same geometric normalization prior to analysis: conversion to 8 bit gray-scale images; rigid scaling and image rotation to place the centers of the two eyes at fixed positions, using the eye coordinates supplied with the original datasets; and image cropping to 120×120 pixels.
Q12. How many pixels of dk give the distance to the nearest image X pixel?
Each pixel of dk gives the distance to the nearest image X pixel with code k (2D Euclidean distance is used in the experiments below).
Q13. What is the way to improve the performance of standard LBP?
Fig. 8 shows the extent to which standard LBP can be improved by combining the three enhancements proposed in this paper: using preprocessing (PP); replacing LBP with LTP; replacing local histogramming and the χ2 histogram distance with the Distance Transform based similarity metric (DT).