Zuoyong Li

Image retrieval based on micro-structure descriptor

This paper presents a simple yet efficient image retrieval approach by proposing a new image feature detector and descriptor, namely the micro-structure descriptor (MSD). The micro-structures are defined based on an edge orientation similarity, and the MSD is built based on the underlying colors in micro-structures with similar edge orientation. With micro-structures serving as a bridge, the MSD extracts features by simulating human early visual processing and it effectively integrates color, texture, shape and color layout information as a whole for image retrieval. The proposed MSD algorithm has high indexing performance and low dimensionality. Specifically, it has only 72 dimensions for full color images, and hence it is very efficient for image retrieval. The proposed method is extensively tested on Corel datasets with 15,000 natural images. The results demonstrate that it is much more efficient and effective than representative feature descriptors, such as Gabor features and multi-textons histogram, for image retrieval.

Content-based image retrieval using computational visual attention model

It is a very challenging problem to well simulate visual attention mechanisms for content-based image retrieval. In this paper, we propose a novel computational visual attention model, namely saliency structure model, for content-based image retrieval. First, a novel visual cue, namely color volume, with edge information together is introduced to detect saliency regions instead of using the primary visual features (e.g., color, intensity and orientation). Second, the energy feature of the gray-level co-occurrence matrices is used for globally suppressing maps, instead of the local maxima normalization operator in Itti?s model. Third, a novel image representation method, namely saliency structure histogram, is proposed to stimulate orientation-selective mechanism for image representation within CBIR framework. We have evaluated the performances of the proposed algorithm on two datasets. The experimental results clearly demonstrate that the proposed algorithm significantly outperforms the standard BOW baseline and micro-structure descriptor. A novel computational visual attention model is developed for content-based image retrieval.Color volume is introduced to detect saliency areas.The energy feature of GLCM is used for globally suppressing map.Orientation-selective mechanism is stimulated for image representation.

Modified directional weighted filter for removal of salt & pepper noise

Switching median filter is a popular type of salt & pepper noise removal technique in recent years. It first detects noise pixels in an image, and then only restores the noise pixels by using the median or its variant of filtering window. Existing directional weighted median filters suffer their own deficiencies when detecting and restoring noise pixels. In this paper, after deeply analyzing the reasons that cause the deficiencies, we propose a modified directional weighted filter to alleviate the issues. The new filter first detects salt & pepper noise by combining existing directional gray level differences with additional judgment of gray level extremes. Then the noise density of each noise pixels non-recursive local window is estimated, and an innovative weighted gray level mean of a recursive or non-recursive filtering window is taken as the restored gray level according to noise density. Experimental results on a series of images show that the proposed algorithm achieves significant improvements in terms of noise suppression and detail preservation, especially when the noise density is high.

Modified local entropy-based transition region extraction and thresholding

Transition region-based thresholding is a newly developed image binarization technique. Transition region descriptor plays a key role in the process, which greatly affects accuracy of transition region extraction and subsequent thresholding. Local entropy (LE), a classic descriptor, considers only frequency of gray level changes, easily causing those non-transition regions with frequent yet slight gray level changes to be misclassified into transition regions. To eliminate the above limitation, a modified descriptor taking both frequency and degree of gray level changes into account is developed. In addition, in the light of human visual perception, a preprocessing step named image transformation is proposed to simplify original images and further enhance segmentation performance. The proposed algorithm was compared with LE, local fuzzy entropy-based method (LFE) and four other thresholding ones on a variety of images including some NDT images, and the experimental results show its superiority.

Robust single-object image segmentation based on salient transition region

Existing transition region-based image thresholding methods are unstable, and fail to achieve satisfactory segmentation accuracy on images with overlapping gray levels between object and background. This is because?they only take the gray level mean of pixels in transition regions as the segmentation threshold of the whole image. To alleviate this issue, we proposed a robust hybrid single-object image segmentation method by exploiting salient transition region. Specifically, the proposed method first uses local complexity and local variance to identify transition regions of an image. Secondly, the transition region with the largest pixel number is chosen as salient transition region. Thirdly, a gray level interval is determined by using transition regions and image information, and one gray level of the interval is determined as the segmentation threshold by using the salient transition region. Finally, the image thresholding result is refined as final segmentation result by using the salient transition region to remove fake object regions. The proposed method has been extensively evaluated by experiments on 170 single-object real world images. Experimental results show that the proposed method achieves better segmentation accuracy and robustness than several types of image segmentation techniques, and enjoys its nature of simplicity and efficiency. We propose a robust salient transition region-based image segmentation method.It adequately uses transition region for image segmentation from a new viewpoint.It alleviates the limitations of transition region-based image thresholding.It significantly improves accuracy and robustness of image segmentation.It keeps the nature of image thresholding, easy operation and high efficiency.

A salt & pepper noise filter based on local and global image information

Existing salt & pepper noise filters only use local image information to detect noise pixels, and neglect global image information. This makes them inapplicable to images with noise-free pixel blocks composed of uncorrupted pixels of gray level extremes, either 0 or 255. In addition, existing filters are hard to simultaneously obtain low miss detection (MD) and low false alarm (FA) in noise detection. To alleviate these issues, we proposed an innovative noise filter based on local and global image information. The proposed filter developed an image block-based method to more accurately estimate noise density of an image, and presented a global image information-based noise detection rectification method. The noise density estimation result was used in subsequent noise detection and rectification stages. Furthermore, the proposed filter combined and slightly revised noise detection schemes of two existing switching filters to improve the accuracy of noise detection. Experimental results on a series of images showed that the proposed filter achieved significant improvement, especially on images with noise-free pixel blocks of gray level extremes.

Illumination-insensitive features for face recognition

Illumination variation is one of the most challenging problems for robust face recognition. In this paper, after investigating the ratio relationship between two neighboring pixels in a digital image, we proposed two illumination-insensitive features, i.e., the non-directional local reflectance normalization (NDLRN) and the fused multi-directional local reflectance normalization (fMDLRN), which not only effectively reduce illumination difference among facial images under different illumination conditions, but also preserve the facial details. Experimental results show that NDLRN and fMDLRN can significantly alleviate the adverse effect of complex illumination on face recognition.

Large cost-sensitive margin distribution machine for imbalanced data classification

This paper develops cost-sensitive margin distribution learning and proposes Large Cost-Sensitive margin Distribution Machine (LCSDM) to get balanced detection rate on imbalanced training data. Recently, margin theory revealed that compared with a single margin, margin distribution is more critical to the generalization performance. Large margin Distribution Machine (LDM) is designed to get superior classification performance and strong generalization performance. However, LDM generally has imbalanced margin distribution between two classes on imbalanced training data. This generally leads to the lower detection rate of the minority class, which contradicts to the needs of high detection rate of the minority class in many real applications. Therefore, cost-sensitive margin distribution learning is brought forward to obtain balanced margin distribution and detection rate between two classes. Whats more, this research deduces the relation between cost-sensitive parameter and in-class detection rate, and designs LCSDM to obtain balanced detection rate. Experimental results show that LCSDM can gradually increase the margin distribution of the minority class to obtain a more balanced detection rate. As a general learning method, LCSDM is especially applicable to imbalanced data classification.

Graph regularized local self-representation for missing value imputation with applications to on-road traffic sensor data

Abstract Recovering missing values (MVs) from incomplete data is an important problem for many real-world applications. Previous research efforts toward solving MVs problem primarily exploit the global and/or local structure of data. In this work, we propose a novel MVs imputation method by combing sample self-representation strategy and underlying local linear structure of data in a uniformed framework. Specifically, the proposed method consists of the following steps. First, an existing method is applied to obtain the first-round estimation of MVs. Then, a graph, characterizing local proximity structure of data, is constructed based on imputed data. Next, a novel model coined as graph regularized local self-representation (GRLSR) is proposed by integrating two crucial elements: local self-representation and graph regularization. The former assumes each sample can be well represented (reconstructed) by linearly combining the neighboring samples while the latter further requires the neighboring samples should not deviate too much from each other after reconstruction. By doing so, MVs can be more accurately restored due to the joint imputation as well as local linear reconstruction. We also develop an effective alternating optimization algorithm to solve GRLSR model, thereby achieving final imputation. The convergence and computational complexity analysis of our method are also presented. To evaluate our method, extensive experiments are conducted on both traffic flow dataset and UCI benchmark datasets. The results demonstrate the effectiveness of our proposed method compared with a set of widely-used competing methods.

Directional Illumination Estimation Sets and Multilevel Matching Metric for Illumination-Robust Face Recognition

It is a challenging task to improve the performance of face recognition under complex illumination conditions. Illumination estimation-based illumination invariant extraction is widely used to alleviate the adverse effects of illumination variation on face recognition. Most existing methods only used slowly changing characteristics of lighting to achieve illumination estimation, thus resulting in inaccurate illumination estimation and illumination invariant extraction under complex illumination conditions. To alleviate this issue, on the basis of the Lambertian reflectance model, we propose an innovative method of directional illumination estimation to extract directional illumination invariant sets from a facial image. The directional illumination invariant sets not only better preserve essential features of the face, but also largely reduce adverse effects of rapid light changes. Moreover, we propose a multilevel matching metric for category classification by using an inner product measure and residual matching. Experimental results on Yale