Lizuo Jin

A hybrid classifier for precise and robust eye detection

Eye location is an important visual cue for face image processing such as alignment before face recognition, gaze tracking, expression analysis, etc. In this paper a novel eye detection algorithm is presented, which integrates the characteristics of single eye and eye-pair images to develop a hybrid classifier under the learning paradigm. The low dimensional features representing eye patterns yield by subspace projection are selected via a filter and a wrapper method for a simplified maximum likelihood and a SVM classifier respectively. Eye candidates determined by a cascade of the two classifiers are further verified with eye-pair template matching scores to reject false detections. The performance of this eye detector is assessed on several publicly available face databases and the experimental results demonstrate its robustness to the variations in head pose, facial expressions, partial occlusions and lighting conditions

Probability estimation for multi-class classification using AdaBoost

Abstract It is a general viewpoint that AdaBoost classifier has excellent performance on classification problems but could not produce good probability estimations. In this paper we put forward a theoretical analysis of probability estimation model and present some verification experiments, which indicate that AdaBoost can be used for probability estimation. With the theory, we suggest some useful measures for using AdaBoost algorithms properly. And then we deduce a probability estimation model for multi-class classification by pairwise coupling. Unlike previous approximate methods, we provide an analytical solution instead of a special iterative procedure. Moreover, a new problem that how to get a robust prediction with classifier scores is proposed. Experiments show that the traditional predict framework, which chooses one with the highest score from all classes as the prediction, is not always good while our model performs well.

Robust urban road image segmentation

Urban road detection with on-board monocular camera in vehicle is still a difficult problem due to its complexity. Combining the superpixel scene segmentation with the classification of texture and structure information of the road scenes, this paper explores the method of detection the road from a single image. Processing the road texture information on a large scale and the structural information around the road on a small scale, and obtaining accurate information about the road edge with scene segmentation, the method is proved to be robust by experiments. Experimental data show that our approach achieved relatively good results in a variety of complex urban environment.

Context-based embedded image compression using binary wavelet transform

Binary wavelet transform (BWT) has several distinct advantages over the real wavelet transform (RWT), such as the conservation of alphabet size of wavelet coefficients, no quantization introduced during the transform and the simple Boolean operations involved. Thus, less coding passes are engaged and no sign bits are required in the compression of transformed coefficients. However, the use of BWT for the embedded grayscale image compression is not well established. This paper proposes a novel Context-based Binary Wavelet Transform Coding approach (CBWTC) that combines the BWT with a high-order context-based arithmetic coding scheme to embedded compression of grayscale images. In our CBWTC algorithm, BWT is applied to decorrelate the linear correlations among image coefficients without expansion of the alphabet size of symbols. In order to match up with the CBWTC algorithm, we employ the gray code representation (GCR) to remove the statistical dependencies among bi-level bitplane images and develop a combined arithmetic coding scheme. In the proposed combined arithmetic coding scheme, three highpass BWT coefficients at the same location are combined to form an octave symbol and then encoded with a ternary arithmetic coder. In this way, the compression performance of our CBWTC algorithm is improved in that it not only alleviate the degradation of predictability caused by the BWT, but also eliminate the correlation of BWT coefficients in the same level subbands. The conditional context of the CBWTC is properly modeled by exploiting the characteristics of the BWT as well as taking advantages of non-causal adaptive context modeling. Experimental results show that the average coding performance of the CBWTC is superior to that of the state-of-the-art grayscale image coders, and always outperforms the JBIG2 algorithm and other BWT-based binary coding technique for a set of test images with different characteristics and resolutions.

Neural network for multi-class classification by boosting composite stumps

We put forward a new model for multi-class classification problems based on the Neural Network structure. The model employs weighted linear regression for feature selection and uses boosting algorithm for ensemble learning. Unlike most previous algorithms, which need to build a collection of binary classifiers independently, the method constructs only one strong classifier once and for all classes via minimizing the total error in a forward stagewise manner. In this work, a novel weak learner framework called composite stump is proposed to improve convergence speed and share features. With these optimization techniques, the classification problem is solved by a simple but effective classifier. Experiments show that the new method outperforms the previous approaches on a number of data sets. HighlightsA novel structure is proposed to improve convergence speed and share features.An adaptive neural network model is presented for multi-class classification.Linear functions are employed as the activation functions in the model.A weighted linear regression with sparsity constraints is used for feature selection.

Online Real Adaboost with Co-training for Object Tracking

One of the major challenges of object tracking is to tackle appearance variations, possibly caused by the change of object postures, size, and occlusions. In this paper an adaptive tracking system is presented, which integrates online semisupervised classification and particle filter efficiently. To identify object pixels from background accurately, classifiers are trained online using real Adaboost which performs much better than its discrete version. In the system, uncorrelated features, color and texture are adopt to train two classifiers separately; the classifiers fused by voting generate confidence score for each pixel measuring its belonging to object or background in candidate regions; accumulated scores in each region are feed to particle filter for estimating object states; pixels with high scores augment the training set mutually and further classifiers are updated by co-training. The system is applied to vehicle and pedestrian tracking in real world scenarios and the experimental results show its robustness to large appearance variations and severe occlusions.

Semi-supervised auto-encoder based on manifold learning.

Auto-encoder is a popular representation learning technique which can capture the generative model of data via a encoding and decoding procedure typically driven by reconstruction errors in an unsupervised way. In this paper, we propose a semi-supervised manifold learning based auto-encoder (named semAE). semAE is based on a regularized auto-encoder framework which leverages semi-supervised manifold learning to impose regularization based on the encoded representation. Our proposed approach suits more practical scenarios in which a small number of labeled data are available in addition to a large number of unlabeled data. Experiments are conducted on several well-known benchmarking datasets to validate the efficacy of semAE from the aspects of both representation and classification. The comparisons to state-of-the-art representation learning methods on classification performance in semi-supervised settings demonstrate the superiority of our approach.

Efficient face recognition fusing dynamic morphological quotient image with local binary pattern

In this paper, we propose a novel illumination normalized Local Binary Pattern (LBP)-based algorithm for face recognition under varying illumination conditions. The proposed DMQI-LBP algorithm fuses illumination normalization, using the Dynamic Morphological Quotient Image (DMQI), into the current LBP-based face recognition system. So it makes full use of advantages of illumination compensation offered by the quotient image, estimated with a dynamic morphological close operation, as well as the powerful discrimination ability provided by the LBP descriptor. Evaluation results on the Yale face database B indicate that the proposed DMQI-LBP algorithm significantly improve the recognition performance (by 5% for the first rank) of the original raw LBP-based system for face recognition with severe lighting variations. Furthermore, our algorithm is efficient and simple to implement, which makes it very suitable for real-time face recognition.

A binary wavelet-based scheme for grayscale image compression

This paper proposes a novel grayscale image compression approach using the binary wavelet transform (BWT) and context-based arithmetic coding, namely the context-based binary wavelet transform coding algorithm (CBWTC). In our CBWTC, in order to alleviate the degradation of predictability caused by the BWT and eliminate the correlation within the same level subbands, three highpass wavelet coefficients at the same location are combined to form an octave symbol and then encoded with a ternary arithmetic coder. The conditional context of the CBWTC is properly modeled by exploiting the properties of the BWT as well as taking the advantages of non-causal adaptive context modeling. Experimental results show that the coding performance of the CBWTC is better than that of the state-of-the-art grayscale image coders except for images containing rich texture, and always outperforms the JBIG2 algorithm and other BWT-based binary coding technique.

Contourlet-based feature extraction for object recognition

A novel contourlet-based local feature descriptor, called Local Contourlet Binary Pattern (LCBP), is developed in this paper. LCBP provides a multiscale and multidirectional representation for images since it integrates contourlet transform with local binary pattern operators. Allowing for the characteristics of marginal and conditional distributions of LCBP as well as simplicity of the model itself, we model LCBP coefficients using a two-state HMT that is in accordance with the intra-band, inter-band and inter-direction distributions of LCBP coefficients. Based on the LCBP-HMT model, we further propose an object recognition method that extracts parameters of the LCBP-HMT model as features and classifies the query sample by comparing the Kullback-Liebler distance between features of the query sample and that of the prototype objects. Experimental results illustrate the superiority of the LCBP over traditional wavelet features and raw statistical features of contourlet coefficients in terms of the discrimination performance.