Jiying Wu

An illumination normalization model for face recognition under varied lighting conditions

In this paper, a novel illumination normalization model is proposed for the pre-processing of face recognition under varied lighting conditions. The novel model could compensate all the illumination effects in face samples, like the diffuse reflection, specular reflection, attached shadow and cast shadow. Firstly, it uses the TV_L^1 model to get the low-frequency part of face image, and adopts the self-quotient model to normalize the diffuse reflection and attached shadow. Then it generates the illumination invariant small-scale part of face sample. Secondly, TV_L^2 model is used to get the noiseless large-scale part of face sample. All kinds of illumination effects in the large-scale part are further removed by the region-based histogram equalization. Thirdly, two parts are fused to generate the illumination invariant face sample. The result of our model contains multi-scaled image information, and all illumination effects in face samples are compensated. Finally, high-order statistical relationships among variables of samples are extracted for classifier. Experimental results on some large scale face databases prove that the processed image by our model could largely improve the recognition performances of conventional methods under low-level lighting conditions.

Independent Gabor Analysis of Discriminant Features Fusion for Face Recognition

A discriminant feature fusion model is proposed for face recognition with large variations of pose, expression, lighting, etc. Discriminant features are extracted by the wavelet transform-based method from two source images. One source image is a holistic gray value image and the other is an illumination invariant geometric image. Face sample is reconstructed by the adaptive fused discriminant feature. Then a bank of Gabor filters is built to extract Gabor representations of the reconstructed samples. Finally higher-order statistical relationships among variables of samples are extracted for classifier. According to experiments, the model outperforms conventional algorithms under complex conditions (large variations of lighting, expression, accessory, etc.).

Independent Gabor Analysis of Multiscale Total Variation-Based Quotient Image

A new algorithm for independent Gabor analysis of multiscale total variation-based quotient image is proposed and applied to face recognition with only one sample per subject here. With our preproposed multiscale TV-based quotient image (TVQI) model, the large-scale and small-scale features are firstly fused to produce the most expressive lighting invariant face. Then a bank of Gabor filters is built to extract lighting invariant Gabor face representations with specified scales and orientations. Last, an information maximization algorithm is adopted to extract higher-order statistical relationships among variables of samples for classifier. According to the experiments on the large-scale CAS-PEAL face database, our approach could outperform Gabor-based ICA, Gabor-based KPCA, and TVQI when they face most outliers (lighting, expression, masking, etc.).

Improved Gradientface used in face recognition under varying illumination

In this paper, an improved Gradientface method is proposed for face recognition under varying illumination. It uses the gradient angle as the input feature. It generates the gradient vectors in difference form, and then computes the gradient angle. The gradient angle which is computed by differential equation preserves the detailed image information and it is proved to be most insensitive to the illumination. Then, the statistical relationships among angles of samples are used for classifier. The cosine distance is the cosine of angle between vectors and then it computes the distance in the gradient domain. Therefore, it achieves the best performance for our method. According to the experimental results, our method outperforms conventional methods under varying illuminations, especially in the large scale face database.

Gabor-based Orthogonal Locality Sensitive Discriminant Analysis for face recognition

An innovative Gabor-based Orthogonal Locality Sensitive Discriminant Analysis for face recognition is presented in this paper. This algorithm is based on a combination of Gabor wavelets representation of face images and a new Orthogonal Locality Sensitive Discriminant Analysis for face recognition. In this paper, a Gabor filter is first designed to extract the features from the whole face images, and then a new Orthogonal Locality Sensitive Discriminant Analysis, which is proposed to preserve the local geometrical structure by computing the mutually orthogonal basis functions iteratively, is used to subject these feature vectors onto locality subspace projection. Experiments based on the ORL face database demonstrate the effectiveness and efficiency of the new method. Results show that our new algorithm is robust to changes in illumination and facial expressions and poses. And it outperforms the other popular approaches reported in the literature and achieves a much higher accurate recognition rate.

Kernel TV-Based Quotient Image Employing Gabor Analysis and Its Application to Face Recognition

In order to overcome the drawback of TVQI and to utilize the property of dimensionality increasing techniques, a novel model for Kernel TV-based Quotient Image employing Gabor analysis is proposed and applied to face recognition with only one sample per subject. To deal with illumination outliers, an enhanced TV-based quotient image (ETVQI) model is first adopted. Then for preprocessed images by ETVQI, a bank of Gabor filters is built to extract features at specified scales and orientations. Lastly, KPCA is introduced to extract final high-order and nonlinear features of extracted Gabor features. According to experiments on the CAS-PEAL face database, our model could outperform Gabor-based KPCA, TVQI and Gabor-based TVQI when they face most outliers (illumination, expression, masking etc.).

A novel image interpolation method based on both local and global information

PDE (Partial differential equation) is an image interpolation method which interpolates based on local geometry property. It can not preserve texture pattern and can only process natural image. NL (Non Local)-means is an interpolation method that uses global information of image. Entire texture pattern in image can be well preserved because of the high replication property of NL-means, while the problem is that blur is preserved as well. In this paper a novel image interpolation method which combines PDE and NL-means is proposed. Image interpolated by the novel method is clear and smooth, and preserves texture pattern. The new method enhances edges using shock filter PDE which does not strengthen jaggies of block contour in interpolated image; the PDE used in this method to smooth image diffuses along level curve. Divided gray regions caused by PDE are smoothed by NL-means; the broken texture pattern is recovered well. Lastly, it is proved that even noisy image can be directly interpolated to the required size using this method. Both theoretical analysis and experiments have been used to verify the benefits of the novel interpolation method.

A novel multi-band image interpolation method

A novel multi-band image interpolation method is proposed. It recovers the final image based on Minimum Mean Square Error (MMSE) rule by using the multi-band image information. Both sub-image features and the multi-band cross-correlated information are used for the estimation. Then we demonstrate the decomposition models which could generate highly-correlated sub-images achieve good performance when adopted in our method. Our method not only considers the properties in sub-images, but incorporates the correlated characteristics among decomposed images in multiple bands, which outperforms conventional single image MMSE estimation methods. Both subjective and objective experimental results prove the validity of our method used in image interpolation.

Multi-scale preprocessing model for face recognition

In this paper a novel multi-scale preprocessing model (MSPM) for face recognition is proposed. MSPM removes lighting effects and enhances the image feature in two scales simultaneously. It decomposes the original image using Total Variation model. Then the lighting effects are normalized by self-quotient in the small scale part and equalized in the large scale part. The final fused image is illumination invariant. Using this image could largely improve face recognition performance under low-level lighting conditions. Combined MSPM with high-order Gabor-based methods could further raise face recognition rates under varying imaging conditions. According to the experiments on the large scale CAS-PEAL face database, MSPM outperforms conventional algorithms when they face most artifacts (lighting, expression, masking etc.).

Gabor-based multi-scale Illumination Normalization model for face recognition

A novel Gabor-based multi-scale illumination normalization (GMSIN) model is proposed and applied to face recognition. GMSIN uses Total variation under different norm constraints. It removes the lighting effect in two scale parts of image and fuses the multi-scaled illumination invariant features. Then a bank of Gabor filters is built to extract lighting invariant Gabor face representations. Finally the higher-order statistical relationships among variables of samples are extracted for classifier. According to the experiments on the large scale CAS-PEAL face database, GMSIN could outperform conventional algorithms when they face most outliers (lighting, expression, masking etc.).