Haibing Ren

LBP Discriminant Analysis for Face Verification

This paper presents a novel Local Binary Pattern (LBP) based Kernel Fisher Discriminant Analysis (KFDA) approach by integrating the LBP descriptor of face images and the KFDA method for face classifier. LBP extracts desirable facial features which consider both shape and texture information to cope with the variation due to facial expression and illumination changes. The KFDA method is then extended to take all the advantages of LBP descriptor for improved face verification performance. We introduce the kernel function by using Chi square statistic distance and RBF as inner product for KFDA classifier. The effectiveness of the LBP based KFDA method with Chi square statistic distance as inner product is shown in terms of comparison with original LBP and KFDA methods on FRGC database.

Subject-independent natural action recognition

A primitive-based dynamic Bayesian networks are proposed for subject-independent natural action recognition. Inferred by high-level knowledge, primitives are distinctive features that describe the context information and the motion information representing human action as well as pose. Dynamic Bayesian networks could fuse multi-information so that many kinds of weak information could function as strong information for inference. The experimental results show that primitive-based dynamic Bayesian networks not only increase the recognition rate but also improve the robustness.

Combining CRF and multi-hypothesis detection for accurate lesion segmentation in breast sonograms

The implementation of lesion segmentation for breast ultrasound image relies on several diagnostic rules on intensity, texture, etc. In this paper, we propose a novel algorithm to achieve a comprehensive decision upon these rules by incorporating image over-segmentation and lesion detection in a pairwise CRF model, rather than a term-by-term translation. Multiple detection hypotheses are used to propagate object-level cues to segments and a unified classifier is trained based on the concatenated features. The experimental results show that our algorithm can avoid the drawbacks of separate detection or bottom-up segmentation, and can deal with very complicated cases.

Multiscale superpixel classification for tumor segmentation in breast ultrasound images

Tumor localization and segmentation in breast ultrasound (BUS) images is an important as well as intractable problem for computer-aided diagnosis (CAD) due to the high variation in shape and appearance. We propose a novel algorithm in this paper without making any assumption on tumor, compared to most previous works. Heterogeneous features are collected via a hierarchical over-segmentation framework, which we have shown has the multiscale property. The superpixels are then classified with their confidences nested into the bottom layer. The ultimate segmentation is made by using an efficient conditional random field model. Experiments on challenging data set show that our algorithm is able to handle almost all kinds of benign and malignant tumors, and also confirm the superiority of our work through a comparison with other two different approaches.

SDTP: a robust method for interest point detection on 3D range images

In fields of intelligent robots and computer vision, the capability to select a few points representing salient structures has always been focused and investigated. In this paper, we present a novel interest point detector for 3D range images, which can be used with good results in applications of surface registration and object recognition. A local shape description around each point in the range image is firstly constructed based on the distribution map of the signed distances to the tangent plane in its local support region. Using this shape description, the interest value is computed for indicating the probability of a point being the interest point. Lastly a Non-Maxima Suppression procedure is performed to select stable interest points on positions that have large surface variation in the vicinity. Our method is robust to noise, occlusion and clutter, which can be seen from the higher repeatability values compared with the state-of-the-art 3D interest point detectors in experiments. In addition, the method can be implemented easily and requires low computation time.

Real-time 3D medical structure segmentation using fast evolving active contours

Segmentation of 3D medical structures in real-time is an important as well as intractable problem for clinical applications due to the high computation and memory cost. We propose a novel fast evolving active contour model in this paper to reduce the requirements of computation and memory. The basic idea is to evolve the brief represented dynamic contour interface as far as possible per iteration. Our method encodes zero level set via a single unordered list, and evolves the list recursively by adding activated adjacent neighbors to its end, resulting in active parts of the zero level set moves far enough per iteration along with list scanning. To guarantee the robustness of this process, a new approximation of curvature for integer valued level set is proposed as the internal force to penalize the list smoothness and restrain the list continual growth. Besides, list scanning times are also used as an upper hard constraint to control the list growing. Together with the internal force, efficient regional and constrained external forces, whose computations are only performed along the unordered list, are also provided to attract the list toward object boundaries. Specially, our model calculates regional force only in a narrowband outside the zero level set and can efficiently segment multiple regions simultaneously as well as handle the background with multiple components. Compared with state-of-the-art algorithms, our algorithm is one-order of magnitude faster with similar segmentation accuracy and can achieve real-time performance for the segmentation of 3D medical structures on a standard PC.

A robust region-based active contour model with point classification for ultrasound breast lesion segmentation

Lesion segmentation is one of the key technologies for computer-aided diagnosis (CAD) system. In this paper, we propose a robust region-based active contour model (ACM) with point classification to segment high-variant breast lesion in ultrasound images. First, a local signed pressure force (LSPF) function is proposed to classify the contour points into two classes: local low contrast class and local high contrast class. Secondly, we build a sub-model for each class. For low contrast class, the sub-model is built by combining global energy with local energy model to find a global optimal solution. For high contrast class, the sub-model is just the local energy model for its good level set initialization. Our final energy model is built by adding the two sub-models. Finally, the model is minimized and evolves the level set contour to get the segmentation result. We compare our method with other state-of-art methods on a very large ultrasound database and the result shows that our method can achieve better performance.

Face recognition using gender information

In this paper, we propose a novel method using gender information for achieving better performances of face recognition systems. Gender is one of the important factors for recognizing appearance of human faces and there are many studies on gender classifications such. However, the gender information is not actively applied in vision-based face recognition tasks, because we cannot find out human identity using only gender information. Therefore, we design the face recognition system based on the gender-based facial features with global facial features, and moreover, gender-based score normalization method for verification task. For fair evaluations, we use FRGC database known as a large size face image database.

Robust vessel detection and segmentation in ultrasound images by a data-driven approach

This paper presents a learning based vessel detection and segmentation method in real-patient ultrasound (US) liver images. We aim at detecting multiple shaped vessels robustly and automatically, including vessels with weak and ambiguous boundaries. Firstly, vessel candidate regions are detected by a data-driven approach. Multi-channel vessel enhancement maps with complement performances are generated and aggregated under a Conditional Random Field (CRF) framework. Vessel candidates are obtained by thresholding the saliency map. Secondly, regional features are extracted and the probability of each region being a vessel is modeled by random forest regression. Finally, a fast levelset method is developed to refine vessel boundaries. Experiments have been carried out on an US liver dataset with 98 patients. The dataset contains both normal and abnormal liver images. The proposed method in this paper is compared with a traditional Hessian based method, and the average precision is promoted by 56 percents and 7.8 percents for vessel detection and classification, respectively. This improvement shows that our method is more robust to noise, therefore has a better performance than the Hessian based method for the detection of vessels with weak and ambiguous boundaries.

3D image noise reduction and contrast enhancement in optical coherence tomography

A novel noise reduction algorithm is proposed for reducing the noise and enhancing the contrast in 3D Optical Coherence Tomography (OCT) images. First, the OCT image is divided into two subregions based on the local noise property: the background area in which the additive noise is dominant and the foreground area in which the multiplicative noise is dominant. In the background, the noise is eliminated by the 2D linear filtering combined with the frame averaging. In the foreground, the noise is eliminated by the 3D linear filtering-an extension of the 2D linear filtering. Therefore, the denoised image is reconstructed according to the combination of the denoised background and foreground. The above procedure can be formulated with a bi-linear model which can be solved efficiently. The proposed bi-linear model can dramatically improve image quality in 3D images with heavy noise and the corresponding linear filter kernel in 2D can be performed in real time. The filter kernel we used is introduced based on the linear noise model in OCT system. The noise model used in the filter kernel includes both the multiplicative (speckle) noise and the additive (incoherent) noise, where the latter is not considered in the most existing linear speckle filters and wavelet filters. Also, the filter kernel can be treated as a low pass filter and can be applied to frequency extraction. Therefore an image contrast enhancement method is introduced in the frequency domain based on the frequency decomposing and weighted combination. A set of experiments are carried out to verify the effectiveness and efficiency of the proposed algorithm.