Limits...
Robust facial expression recognition via compressive sensing.

Zhang S, Zhao X, Lei B - Sensors (Basel) (2012)

Bottom Line: The CS theory is used to construct a sparse representation classifier (SRC).The effectiveness and robustness of the SRC method is investigated on clean and occluded facial expression images.Three typical facial features, i.e., the raw pixels, Gabor wavelets representation and local binary patterns (LBP), are extracted to evaluate the performance of the SRC method.

View Article: PubMed Central - PubMed

Affiliation: School of Physics and Electronic Engineering, Taizhou University, Taizhou 318000, China. tzczsq@163.com

ABSTRACT
Recently, compressive sensing (CS) has attracted increasing attention in the areas of signal processing, computer vision and pattern recognition. In this paper, a new method based on the CS theory is presented for robust facial expression recognition. The CS theory is used to construct a sparse representation classifier (SRC). The effectiveness and robustness of the SRC method is investigated on clean and occluded facial expression images. Three typical facial features, i.e., the raw pixels, Gabor wavelets representation and local binary patterns (LBP), are extracted to evaluate the performance of the SRC method. Compared with the nearest neighbor (NN), linear support vector machines (SVM) and the nearest subspace (NS), experimental results on the popular Cohn-Kanade facial expression database demonstrate that the SRC method obtains better performance and stronger robustness to corruption and occlusion on robust facial expression recognition tasks.

No MeSH data available.


Related in: MedlinePlus

A corrupted image example (a) Original image of 640 × 490 pixels; (b) Resized image of 32 × 32 pixels; (c) 50% corrupted image.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3376615&req=5

f5-sensors-12-03747: A corrupted image example (a) Original image of 640 × 490 pixels; (b) Resized image of 32 × 32 pixels; (c) 50% corrupted image.

Mentions: At first, we consider the recognition of facial expressions with different percentage of image pixels corrupted at random. The percentage of the pixels are randomly chosen from each of test image and replaced by random values in the range [0, Mi], where Mi is the maximum pixel value in the ith test image. The percentage of corrupted pixels varies from 0% to 90%. Figure 5 gives an example of a 50% corrupted face image on the resized image of 32 × 32 pixels. As shown in Figure 5, beyond 50% corruption, the corrupted images are scarcely identified as facial images. Figure 6 plots the recognition accuracy of all used methods, i.e., NN, SVM, NS and SRC, under different percentage corrupted from 0% to 90%. It can be observed that the performance of all used methods decreased as the percentage corrupted increased. Nevertheless, SRC still dramatically outperforms the other used methods at various levels of corruption. This indicates SRC is more robust to the random pixels corruption than the other used methods.


Robust facial expression recognition via compressive sensing.

Zhang S, Zhao X, Lei B - Sensors (Basel) (2012)

A corrupted image example (a) Original image of 640 × 490 pixels; (b) Resized image of 32 × 32 pixels; (c) 50% corrupted image.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3376615&req=5

f5-sensors-12-03747: A corrupted image example (a) Original image of 640 × 490 pixels; (b) Resized image of 32 × 32 pixels; (c) 50% corrupted image.
Mentions: At first, we consider the recognition of facial expressions with different percentage of image pixels corrupted at random. The percentage of the pixels are randomly chosen from each of test image and replaced by random values in the range [0, Mi], where Mi is the maximum pixel value in the ith test image. The percentage of corrupted pixels varies from 0% to 90%. Figure 5 gives an example of a 50% corrupted face image on the resized image of 32 × 32 pixels. As shown in Figure 5, beyond 50% corruption, the corrupted images are scarcely identified as facial images. Figure 6 plots the recognition accuracy of all used methods, i.e., NN, SVM, NS and SRC, under different percentage corrupted from 0% to 90%. It can be observed that the performance of all used methods decreased as the percentage corrupted increased. Nevertheless, SRC still dramatically outperforms the other used methods at various levels of corruption. This indicates SRC is more robust to the random pixels corruption than the other used methods.

Bottom Line: The CS theory is used to construct a sparse representation classifier (SRC).The effectiveness and robustness of the SRC method is investigated on clean and occluded facial expression images.Three typical facial features, i.e., the raw pixels, Gabor wavelets representation and local binary patterns (LBP), are extracted to evaluate the performance of the SRC method.

View Article: PubMed Central - PubMed

Affiliation: School of Physics and Electronic Engineering, Taizhou University, Taizhou 318000, China. tzczsq@163.com

ABSTRACT
Recently, compressive sensing (CS) has attracted increasing attention in the areas of signal processing, computer vision and pattern recognition. In this paper, a new method based on the CS theory is presented for robust facial expression recognition. The CS theory is used to construct a sparse representation classifier (SRC). The effectiveness and robustness of the SRC method is investigated on clean and occluded facial expression images. Three typical facial features, i.e., the raw pixels, Gabor wavelets representation and local binary patterns (LBP), are extracted to evaluate the performance of the SRC method. Compared with the nearest neighbor (NN), linear support vector machines (SVM) and the nearest subspace (NS), experimental results on the popular Cohn-Kanade facial expression database demonstrate that the SRC method obtains better performance and stronger robustness to corruption and occlusion on robust facial expression recognition tasks.

No MeSH data available.


Related in: MedlinePlus