Lansun Shen

Optimal sampling of Gabor features for face recognition

The Gabor feature is effective for facial image representation, while linear discriminant analysis (LDA) can extract the most discriminant information from the Gabor feature for face recognition. In practice, the dimension of a Gabor feature vector is so high that the computation and memory requirements are prohibitively large. To reduce the dimension, one simple scheme is to extract the Gabor feature at sub-sampled positions, usually in a regular grid, in a face region. However, this scheme is not effective enough and degrades the recognition performance. In this paper, we propose a method to determine the optimal position for extracting the Gabor feature such that the number of feature points is as small as possible while the representation capability of the points is as high as possible. The subsampled positions of the feature points are determined by a mask generated from a set of training images by means of principal component analysis (PCA). With the feature vector of reduced dimension, a subspace LDA is applied for face recognition, i.e., PCA is first used to reduce the dimension of the Gabor feature vectors generated from the subsampled positions, and then a common LDA is applied. Experimental results show that the new sampling method is simple, and effective for both dimension reduction and image representation. The recognition rate based on our proposed scheme is also higher than that achieved using a regular sampling method in a face region.

Illumination invariant face recognition

The appearance of a face will vary drastically when the illumination changes. Variations in lighting conditions make face recognition an even more challenging and difficult task. In this paper, we propose a novel approach to handle the illumination problem. Our method can restore a face image captured under arbitrary lighting conditions to one with frontal illumination by using a ratio-image between the face image and a reference face image, both of which are blurred by a Gaussian filter. An iterative algorithm is then used to update the reference image, which is reconstructed from the restored image by means of principal component analysis (PCA), in order to obtain a visually better restored image. Image processing techniques are also used to improve the quality of the restored image. To evaluate the performance of our algorithm, restored images with frontal illumination are used for face recognition by means of PCA. Experimental results demonstrate that face recognition using our method can achieve a higher recognition rate based on the Yale B database and the Yale database. Our algorithm has several advantages over other previous algorithms: (1) it does not need to estimate the face surface normals and the light source directions, (2) it does not need many images captured under different lighting conditions for each person, nor a set of bootstrap images that includes many images with different illuminations, and (3) it does not need to detect accurate positions of some facial feature points or to warp the image for alignment, etc.

Example-based image super-resolution with class-specific predictors

Example-based super-resolution is a promising approach to solving the image super-resolution problem. However, the learning process can be slow and prediction can be inaccurate. In this paper, we present a novel learning-based algorithm for image super-resolution to improve the computational speed and prediction accuracy. Our new method classifies image patches into several classes, for each class, a class-specific predictor is designed. A class-specific predictor takes a low-resolution image patch as input and predicts a corresponding high-resolution patch as output. The performances of the class-specific predictors are evaluated using different datasets formed by face images and natural-scene images. We present experimental results which demonstrate that the new method provides improved performances over existing methods.

An adaptive algorithm for the display of high-dynamic range images

A novel algorithm based on spatial and statistical information is proposed for the display of high-dynamic range (HDR) images. In our proposed algorithm, an image is first decomposed into a base layer and a detailed layer, which represent its smoothed and fine details, respectively. The problem of overall impression preservation is regarded as a global issue in our algorithm. Statistical-based histogram adjustment is employed to deal with the base layer. The reproduction of visual details is regarded as a local issue. The detailed layer obtained using a spatial filter is adaptively enhanced according to the mapping function used for the base layer. The main contributions of our algorithm are that: (1) an adaptive detail-enhancement method is proposed and (2) a gain map is defined to combine the local and global issues. Experimental results show the superior performance of our approach in terms of visual quality.

Optimization and Implementation of H.264 Encoder on DSP Platform

Compared with MPEG-4 and other previous standards, H.264 standard has achieved great breakthrough in coding performance. In this paper, the optimization and implementation of H.264 baseline profile encoder on the TMS320DM642 has been presented. Based on the architectural features of TMS320DM642 and the computational complexity analysis of H.264 encoder, the H.264 encoder has been optimized from three aspects: algorithms, data transfer and memory/Cache use. The experimental results demonstrate that, for the video sequences with CIF format, the optimized H.264 encoder can achieve the encoding speed of more than 24 frames per second, which can meet the real-time requirements of the applications.

Regions of Interest extraction based on visual attention model and watershed segmentation

The presented research addressed a novel visual attention model and watershed segmentation based approach of regions of interest (ROIs) extraction, which automatically extracts ROIs and copes with the watershed over-segmentation. This approach uses visual attention model to locate salient points, in which the winner point, the most salient point, is selected as the seed point of watershed segmentation. ROIs are extracted by combining salient regions with watershed segmented regions. The focus of attention (FOA) is shifted to measure the importance or interest of the extracted regions. The experimental results show that the proposed method is effective to reduce over-segmentation in auto-extracting ROIs and performs well for different objects.

An efficient example-based approach for image super-resolution

A novel algorithm for image super-resolution with class-specific predictors is proposed in this paper. In our algorithm, the training example images are classified into several classes, and each patch of a low-resolution image is classified into one of these classes. Each class has its high-frequency information inferred using a class-specific predictor, which is trained via the training samples from the same class. In this paper, two different types of training sets are employed to investigate the impact of the training database to be used. Experimental results have shown the superior performance of our method.

A PERSONALIZED IMAGE RETRIEVAL BASED ON USER INTEREST MODEL

In order to narrow the semantic gap, user interest model plays an important role in personalized image retrieval. A novel personalized image retrieval approach based on user interest model is proposed in this study. User interest model is developed on the basis of short-tem and long-term interests. (1) Short-term interests are represented by collecting visual and semantic features. Visual features are collected by MARS relevance feedback. Semantic features are constructed by building a mapping from image low-level visual features to high-level semantic features on the basis of SVM. (2) Long-term interests are inferred by inference engine from the collected short-term interests. Long-term visual features are collected by the nonlinear gradual forgetting interest inference algorithm and semantic features are obtained by clustering algorithm. After applying to image retrieval, experimental results show that the average recall/precision is significantly improved and a better user satisfaction rate is achieved as well. Furthermore, it demonstrates our model can be efficiently adapted to user interests and matches personalized image retrieval.

Complexity scalable control for H.264 motion estimation and mode decision under energy constraints

The H.264 video coding standard supports several inter-prediction coding modes using variable block sizes. The robust rate-distortion optimization (RDO) technique of both the motion estimation (ME) and the mode decision (MD) is adopted to achieve superior coding efficiency, which also entails a lot of complex computations. In this paper, by rearranging the order of candidate modes and using two new rate-distortion-complexity optimization (RDCO) functions, a more efficient RDCO framework for fast ME and MD is first proposed. Then, with this new framework, a novel complexity scalable control algorithm for H.264 inter-prediction coding is devised, which is based on an efficient complexity allocation and control (CAAC) scheme performed both at the frame level and the macroblock (MB) level. Experimental results show that our proposed algorithm can adaptively determine an appropriate cut-off point for a sequence of re-ordered candidate modes according to the current energy condition of a mobile device, which can adjust the complexity of inter-prediction coding to any appropriate level with minimum degradation in video quality. This can therefore prolong the operational lifetime of batteries.

A Data-Mining Based Skin Detection Method in JPEG Compressed Domain

A novel skin detection method in JPEG compressed domain has been proposed in this paper. Color and texture features of the image blocks are extracted from the entropy decoded DCT coefficients firstly. Then, data mining method, i.e. decision tree, is applied to establish the skin color model to describe the relationship between the features of image blocks and the skin detection results, afterwards, the initial skin regions are detected based on the skin color model. The skin regions are finally, segmented through region growing method. Experimental results show that, compared with the SPM (Skin Probability Map) skin detection method in the pixel domain, the proposed method can achieve higher detection accuracy as well as higher speed.