Quality assessment based denoising to improve face recognition performance
20 Jun 2011-pp 140-145
TL;DR: In the proposed framework, the optimal parameter set is selected for denoising using quality assessment algorithms with low complexity using quality score based parameter selection on the AR face dataset.
Abstract: A probe face image may contain noise due to environmental conditions, incorrect use of sensors or transmission error. The performance of face recognition severely depletes when the probe image is contaminated with noise. Denoising techniques can improve recognition performance, provided the correct parameters are used. In this paper, a parameter selection framework is presented. In the proposed framework, the optimal parameter set is selected for denoising using quality assessment algorithms with low complexity. Quality score based parameter selection is evaluated on the AR face dataset. A correlation study is discussed to ascertain the relationship between the quality scores and recognition rate. The experiments suggest that the proposed framework improves the performance both in terms of accuracy and computation time.
Citations
More filters
27 Jan 2015
TL;DR: A comprehensive survey of current advancements in quality assessment in biometrics and recent advances and applicability of quality in multibiometrics is presented, and quality in face recognition is explored, an area yet to receive proportionate attention from the research community.
Abstract: Quality is an attribute or a property of an item that quantita tively measures specific aspect or content. The definition an d correct method of measurement of quality of a biometric modality tha t is usually represented by an image is currently unclear in t he research community. While a biometric image’s quality is su ceptible to degradation during capture and storage, it may also have low quality by its very nature. Quality of a biometric has sev eral applications in popular research interests such as i) u nconstrained biometric recognition ii) multibiometrics and iii) largescale identity projects. This research aims to define and demystify quality in the field of biometrics. We present a comprehensive survey of current advancements in quality assessment, starting with a concis e summary of the field ofBiometrics and recent advances and applicability of quality inmultibiometrics. In order to understand quality assessment in biometrics, w e delve into related area of image quality assessment. Further, several applications a nd factors that influence biometric quality are analyzed. We also investigate popular methods of evaluating quality assessment algorith ms in biometrics. Finally, we explore quality in face recogn itio , an area that is yet to receive proportionate attention from the research community. The complexity of the problem is multip lied by the lack of consensus in literature on the definition and cons titution of facial features. However, initial experiments indicate that holistic image descriptors are able to successfully encode degradations in biometric images.
6 citations
Cites background from "Quality assessment based denoising ..."
...Framework for a) a quality driven biome tric image enhancement, based on [71] and b) quality based multi-classifier selection, propo sed by [54] ....
[...]
...Denoising techniques help in improving the recognizability of face images, provided the correct parameter are used [71]....
[...]
TL;DR: It is shown that MVQ can be effectively used to control the relative performance of recall and precision that can be achieved on enhanced images and for certain face detection algorithms, optimizing for MVQ instead of perceptual quality can lead to improved face detection performance.
Abstract: Machine vision quality (MVQ) assessment for face detection refers to image quality as judged by face detection algorithms. While perceptual image quality usually depends on human observers, MVQ depends on factors such as the face detection algorithm and performance measures such as recall and precision. We define the MVQ index as a weighted combination of the predicted precision and recall of the face detection algorithm and predict it using image features based on natural scene statistics. A filter bank framework is developed to achieve image enhancement of distorted images for face detection, where the enhancement operations can be optimized for either perceptual quality or MVQ. It is shown that MVQ can be effectively used to control the relative performance of recall and precision that can be achieved on enhanced images. For certain face detection algorithms, optimizing for MVQ instead of perceptual quality can lead to improved face detection performance. The MVQ is developed for three different face detection algorithms and the image enhancement framework is tested on a dataset based on the IDEAL-LIVE Distorted Face Database. A computationally efficient method to optimize for MVQ is designed by predicting the MVQ of enhanced images directly from features extracted from distorted images and filter parameters. The optimization framework is further enhanced by allowing for filter bank independent optimization of MVQ.
6 citations
Cites methods from "Quality assessment based denoising ..."
...Parameters of denoising algorithms have been tuned to improve face recognition performance by using low complexity image quality assessment algorithms [15]....
[...]
TL;DR: A facial image verification and quality assessment framework (FaceIVQA) was developed that produced accurate and consistent facial image assessment data and can be used as an input to quality-driven biometric fusion systems.
Abstract: Although several techniques have been proposed for predicting biometric system performance using quality values, many of the research works were based on no-reference assessment technique using a single quality attribute measured directly from the data. These techniques have proved to be inappropriate for facial verification scenarios and inefficient because no single quality attribute can sufficient measure the quality of a facial image. In this research work, a facial image verification and quality assessment framework (FaceIVQA) was developed. Different algorithms and methods were implemented in FaceIVQA to extract the faceness, pose, illumination, contrast and similarity quality attributes using an objective full-reference image quality assessment approach. Structured image verification experiments were conducted on the surveillance camera (SCface) database to collect individual quality scores and algorithm matching scores from FaceIVQA using three recognition algorithms namely principal component analysis (PCA), linear discriminant analysis (LDA) and a commercial recognition SDK. FaceIVQA produced accurate and consistent facial image assessment data. The Result shows that it accurately assigns quality scores to probe image samples. The resulting quality score can be assigned to images captured for enrolment or recognition and can be used as an input to quality-driven biometric fusion systems. DOI: http://dx.doi.org/10.11591/ijece.v3i6.5034
6 citations
Cites background from "Quality assessment based denoising ..."
...Several researchers have made attempts to measure biometric system performance using image quality assessment and prediction but many of these research works were based on no-reference quality assessment techniques and the assessment evaluation is usually focused on the biometric samples themselves, thereby using quality measures directly calculated from the data, such as denoising techniques [13], the signal-to-noise-ratio [14], similarity surface analysis [15], modelling recognition similarity scores [6], high frequency components of discrete cosine transformation [16], difference in image intensity [17] and image activity estimation in both horizontal and vertical direction [18]....
[...]
TL;DR: Experimental results showed that the performance of face recognition using the proposed method was better than that of conventional methods in terms of accuracy.
Abstract: With the rapid growth of smart TV, the necessity for recognizing a viewer has increased for various applications that deploy face recognition to provide intelligent services and high convenience to viewers. However, the viewers can have various postures, illumination, and expression variations on their faces while watching TV, and thereby, the performance of face recognition inevitably degrades. In order to handle these problems, video-based face recognition has been proposed, instead of a single image-based one. However, video-based processing of multiple images is prohibitive in smart TVs as the processing power is limited. Therefore, a quality measure-based (QM-based) image selection is required that considers both the processing speed and accuracy of face recognition. Therefore, we propose a performance enhancement method for face recognition through symmetrical fuzzy-based quality assessment. Our research is novel in the following three ways as compared to previous works. First, QMs are adaptively selected by comparing variance values obtained from candidate QMs within a video sequence, where the higher the variance value by a QM, the more meaningful is the QM in terms of a distinction between images. Therefore, we can adaptively select meaningful QMs that reflect the primary factors influencing the performance of face recognition. Second, a quality score of an image is calculated using a fuzzy method based on the inputs of the selected QMs, symmetrical membership functions, and rule table considering the characteristics of symmetry. A fuzzy-based combination method of image quality has the advantage of being less affected by the types of face databases because it does not perform an additional training procedure. Third, the accuracy of face recognition is enhanced by fusing the matching scores of the high-quality face images, which are selected based on the quality scores among successive face mages. Experimental results showed that the performance of face recognition using the proposed method was better than that of conventional methods in terms of accuracy.
5 citations
01 Jan 2013
TL;DR: A comparison between face recognition rate with noise and face Recognition rate without noise is presented and the proposed Haar10 method on PCA, Linear Discriminate Analysis (LDA), Kernel PC a, Fisher Analysis (FA) face recognition methods is implemented.
Abstract: In this paper a comparison between face recognition rate with noise and face recognition rate without noise is presented. In our work we assume that all the images in the ORL faces database are noisy images. We applied the wavelet based image de-noising methods to this database and created new databases, then the face recognition rate are calculated to them. Three experiments are given in our paper. In the first experiment different wavelet methods with different level of decomposition (up to ten decompositions) are used for de-noising the ORL database and the comparison is done when Principal Components Analysis (PCA) is applied to evaluate the verification rate. In the second experiment de-noising different sets of ORL database with methods that have best performance in levels (1, 2, 3, and 10) is done (as a result from experiment 1). In the third experiment we implement the proposed Haar10 method on PCA, Linear Discriminate Analysis (LDA), Kernel PCA, Fisher Analysis (FA) face recognition methods and the recognition rates are evaluated for both the noisy and de-noisy databases.
5 citations
References
More filters
TL;DR: In this article, a structural similarity index is proposed for image quality assessment based on the degradation of structural information, which can be applied to both subjective ratings and objective methods on a database of images compressed with JPEG and JPEG2000.
Abstract: Objective methods for assessing perceptual image quality traditionally attempted to quantify the visibility of errors (differences) between a distorted image and a reference image using a variety of known properties of the human visual system. Under the assumption that human visual perception is highly adapted for extracting structural information from a scene, we introduce an alternative complementary framework for quality assessment based on the degradation of structural information. As a specific example of this concept, we develop a structural similarity index and demonstrate its promise through a set of intuitive examples, as well as comparison to both subjective ratings and state-of-the-art objective methods on a database of images compressed with JPEG and JPEG2000. A MATLAB implementation of the proposed algorithm is available online at http://www.cns.nyu.edu//spl sim/lcv/ssim/.
40,609 citations
TL;DR: A generalized gray-scale and rotation invariant operator presentation that allows for detecting the "uniform" patterns for any quantization of the angular space and for any spatial resolution and presents a method for combining multiple operators for multiresolution analysis.
Abstract: Presents a theoretically very simple, yet efficient, multiresolution approach to gray-scale and rotation invariant texture classification based on local binary patterns and nonparametric discrimination of sample and prototype distributions. The method is based on recognizing that certain local binary patterns, termed "uniform," are fundamental properties of local image texture and their occurrence histogram is proven to be a very powerful texture feature. We derive a generalized gray-scale and rotation invariant operator presentation that allows for detecting the "uniform" patterns for any quantization of the angular space and for any spatial resolution and presents a method for combining multiple operators for multiresolution analysis. The proposed approach is very robust in terms of gray-scale variations since the operator is, by definition, invariant against any monotonic transformation of the gray scale. Another advantage is computational simplicity as the operator can be realized with a few operations in a small neighborhood and a lookup table. Experimental results demonstrate that good discrimination can be achieved with the occurrence statistics of simple rotation invariant local binary patterns.
14,245 citations
"Quality assessment based denoising ..." refers background or methods in this paper
...ture algorithms such as Local Binary Patterns (LBP) [7], are known to be more resilient towards these covariates compared to appearance based algorithms such as Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA)....
[...]
...The experiment performed using data-driven noise and LBP demonstrates the effect of noise on face recognition....
[...]
...The recognition accuracy is also computed for each set using LBP based face recognition algorithm [7]....
[...]
...∙ The class label corresponding to each of the quality vector is the parameter 𝑃1..𝑖 which results in the best rank-1 efficiency with the training-gallery-set using local binary pattern (LBP) as the face recognition algorithm....
[...]
...Tex- ture algorithms such as Local Binary Patterns (LBP) [7], are known to be more resilient towards these covariates compared to appearance based algorithms such as Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA)....
[...]
[...]
01 Jan 1998
3,650 citations
"Quality assessment based denoising ..." refers methods in this paper
...The experiments are conducted on the AR face database [5] containing 756 frontal face images pertaining to 126 subjects (i....
[...]
...As discussed in Section 4, SVM model is learned using the training labels from the data driven approach on the training set of 50 individuals from the AR face dataset[5]....
[...]
Journal Article•
2,952 citations
"Quality assessment based denoising ..." refers methods in this paper
...As discussed in Section 4, SVM model is learned using the training labels from the data driven approach on the training set of 50 individuals from the AR face dataset[5]....
[...]
...The experiments are conducted on the AR face database [5] containing 756 frontal face images pertaining to 126 subjects (i....
[...]
TL;DR: An adaptive, data-driven threshold for image denoising via wavelet soft-thresholding derived in a Bayesian framework, and the prior used on the wavelet coefficients is the generalized Gaussian distribution widely used in image processing applications.
Abstract: The first part of this paper proposes an adaptive, data-driven threshold for image denoising via wavelet soft-thresholding. The threshold is derived in a Bayesian framework, and the prior used on the wavelet coefficients is the generalized Gaussian distribution (GGD) widely used in image processing applications. The proposed threshold is simple and closed-form, and it is adaptive to each subband because it depends on data-driven estimates of the parameters. Experimental results show that the proposed method, called BayesShrink, is typically within 5% of the MSE of the best soft-thresholding benchmark with the image assumed known. It also outperforms SureShrink (Donoho and Johnstone 1994, 1995; Donoho 1995) most of the time. The second part of the paper attempts to further validate claims that lossy compression can be used for denoising. The BayesShrink threshold can aid in the parameter selection of a coder designed with the intention of denoising, and thus achieving simultaneous denoising and compression. Specifically, the zero-zone in the quantization step of compression is analogous to the threshold value in the thresholding function. The remaining coder design parameters are chosen based on a criterion derived from Rissanen's minimum description length (MDL) principle. Experiments show that this compression method does indeed remove noise significantly, especially for large noise power. However, it introduces quantization noise and should be used only if bitrate were an additional concern to denoising.
2,917 citations
"Quality assessment based denoising ..." refers background or methods in this paper
...The training labels for the parameter selection are 1For further details of BayesShrink, refer to Chang et al.[1]....
[...]
...In the proposed quality assessment based denoising framework, wavelet based soft thresholding technique is used for denoising, also known as BayesShrink [1]....
[...]
...∙ Each of these corrupt training-probe-set are denoised with the wavelet based BayesShrink denoising algorithm[1] with each of the 𝑖 candidate parameters 𝑃1..𝑖. ∙ The quality vector for each image with quality scores [𝑄1, 𝑄2] is computed and used as the training sample for a multi-class SVM classifier....
[...]
...Several enhancement methods have been proposed in literature to handle these corruptions [1]....
[...]
...∙ Each of these corrupt training-probe-set are denoised with the wavelet based BayesShrink denoising algorithm[1] with each of the i candidate parameters P1....
[...]