Showing papers by "Xiangjian He published in 2008"

Segmentation of characters on car license plates

Abstract: License plate recognition usually contains three steps, namely license plate detection/localization, character segmentation and character recognition. When reading characters on a license plate one by one after license plate detection step, it is crucial to accurately segment the characters. The segmentation step may be affected by many factors such as license plate boundaries (frames). The recognition accuracy will be significantly reduced if the characters are not properly segmented. This paper presents an efficient algorithm for character segmentation on a license plate. The algorithm follows the step that detects the license plates using an AdaBoost algorithm. It is based on an efficient and accurate skew and slant correction of license plates, and works together with boundary (frame) removal of license plates. The algorithm is efficient and can be applied in real-time applications. The experiments are performed to show the accuracy of segmentation.

Human Action Recognition by Radon Transform

Abstract: A new feature description is used for human behaviour representation and recognition. The feature is based on Radon transforms of extracted silhouettes. Key postures are selected based on the Radon transform. Key postures are combined to construct an action template for each sequence. Linear discriminant analysis (LDA) is applied to the set of key postures to obtain low dimensional feature vectors. Different classification methods are used to classify each sequence. Experiments are carried out based on a publically available human behaviour database and the results are exciting.

An approach of canny edge detection with virtual hexagonal image structure

Abstract: Edge detection plays an important role in the areas of image processing, multimedia and computer vision. Gradient-based edge detection is a straightforward method to identify the edge points in the original grey-level image. It is intuitive that, in the human vision system, the edge points always appear where the gradient magnitude assumes a maximum. Hexagonal structure is an image structure alternative to traditional square image structure. The geometrical arrangement of pixels on a hexagonal structure can be described as a collection of hexagonal pixels. Because all the existing hardware for capturing image and for displaying image are produced based on square structure, an approach that uses bilinear interpolation and tri-linear interpolation is applied for conversion between square and hexagonal structures. Based on this approach, an edge detection method is proposed. This method performs Gaussian filtering to suppress image noise and computes gradients on the hexagonal structure. The pixel edge strengths on the square structure are then estimated before Canny's edge detector is applied to determine the final edge map. The experimental results show that the proposed method improves the edge detection accuracy and efficiency.

Using dynamic programming to match human behavior sequences

Abstract: This paper proposed a new approach for recognition and matching the human behavior sequence. Each human behavior sequence is represented by its key postures to greatly reduce the computation time. Normalization is applied to all the behavior sequences key postures for matching. A dynamic time warping (DTW) algorithm is used to perform the alignment of two time series. Experiments are carried out on an open human behavior database and exciting results have been obtained.

Motion Based Pedestrian Recognition

Abstract: This paper proposed a method for discriminating pedestrians from rigid objects in a video. The methodis a motion-based recognition of moving objects. This method is motivated by the assumptions that human beings are non-rigid and their movements are periodic. Moving objects and their skeletons are extracted. The motion cue is determined by the angle formed by the centroid point and the two bottom end points at object’s skeleton. The histogram of the cue over a time period is used to determine if the object is pedestrian or not. This cue does not require any pre-built models. Neither does it need Fourier Transform to obtain the cycle of the objects. The proposed method is computation inexpensive, and it can be used for realtime video surveillance.

A hierarchically combined classifier for license plate recognition

Abstract: High accuracy and fast recognition speed are two requirements for real-time and automatic license plate recognition system. In this paper, we propose a hierarchically combined classifier based on an inductive learning based method and an SVM-based classification. This approach employs the inductive learning based method to roughly divide all classes into smaller groups. Then the SVM method is used for character classification in individual groups. Both start from a collection of samples of characters from license plates. After a training process using some known samples in advance, the inductive learning rules are extracted for rough classification and the parameters used for SVM-based classification are obtained. Then, a classification tree is constructed for further fast training and testing processes for SVM-based classification. Experimental results for the proposed approach are given. From the experimental results, we can make the conclusion that the hierarchically combined classifier is better than either the inductive learning based classification or the SVM-based classification in terms of error rates and processing speeds.

Extracting key postures in a human action video sequence

Abstract: Human key posture extraction from videos will benefit video storage, video retrieval, human action recognition, human behaviour understanding and so on. This paper presents an approach to select key postures from human action sequences using 2D information. There are two steps in the proposed method. Information measurement which is a kind of global feature of a frame is used to roughly find key posture candidates. Then, a body skeleton feature which is a kind of local feature is applied to select final key postures from the candidates obtained in the first step. The experiments show that the proposed method is efficient.

Local Binary Patterns with Mahalanobis Distance Maps for Human Detection

Abstract: Local Binary Pattern (LBP) was designed and has been widely used for efficient texture classification. LBP provides a simple and effective way to represent texture patterns. Uniform LBPs play an important role for LBP-based pattern/object recognition as they include majority of LBPs. On the other hand, Human detection based on Mahalanobis Distance Map (MDM) recognizes appearance of human based on geometrical structure. Each MDM shows a clear texture pattern that can be classified using LBPs. In this paper, we compute LBPs of MDMs. Chi-square as a measure is used for human detection based on uniform LBPs obtained. We show that our method using LBPs built on MDMs has a higher human detection rate and a lower false positive rate compared to the method merely based on MDMs.

Edge Detection on Hexagonal Structure

Abstract: Gradient-based edge detection is a straightforward method to identify the edge points in the original grey-level image. It is intuitive that in the human vision system the edge points always appear where the grey-level value is greatly changed. Spiral Architecture is a relatively new image data structure that is inspired from anatomical considerations of the primate's vision. In Spiral Architecture, each image is represented as a collection of hexagonal pixels. Edge detection on Spiral Architecture has features of fast computation and accurate localization. In this paper, we present and compare gradient-based edge detection algorithms on Spiral Architecture. The experimental results show that the edge detection on Spiral Architecture outperforms that on traditional square image structure.

A modified Mahalanobis distance for human detection in out-door environments

Abstract: This paper proposes a novel method for human detection from static images based on pixel structure of input images. Each image is divided into four parts, and a weight is assigned to each part of the image. In training stage, all sample images including human images and non-human images are used to construct a Mahalanobis distance map through statistically analyzing the difference between the different blocks on each original image. A projection matrix will be created with Linear Discriminant Method (LDM) based on the Mahalanobis distance map. This projection matrix will be used to transform multi-dimensional feature vectors into one dimensional feature domain according to a pre-calculated threshold to distinguish human figures from non-human figures. In comparison with the method without introducing weights, the proposed method performs much better. Encouraging experimental results have been obtained based on MIT dataset and our own dataset.

A bank customer credit evaluation based on the decision tree and the simulated annealing algorithm

Abstract: C4.5 is a learning algorithm that adopts local search strategy, and it cannot obtain the best decision rules. On the other hand, the simulated annealing algorithm is a globally optimized algorithm and it avoids the drawbacks of C4.5. This paper proposes a new credit evaluation method based on decision tree and simulated annealing algorithm. The experimental results demonstrate that the proposed method is effective.

Fractal Image Compression on Hexagonal Structure

Abstract: Fractal image compression (FIC) is a relatively recent image compression method. Its basic idea is to represent images as a fixed point of a contractive Iterated Function System (IFS). Spiral Architecture (SA) is a novel hexagonal image structure on which images are displayed as a collection of hexagonal pixels. The efficiency and accuracy of image processing on SA have been demonstrated in many recently published papers. In this paper, two presentations of SA on the traditional display device will be discussed. Then we will review the current research work on fractal image compression based on SA using both presentations. The FIC performance on SA will be compared with it on the traditional square structure in terms of compression ratio and PSNR. In the experimental results, higher PSNR values can be achieved at various compression ratios for all test images. The preliminary research on this direction has shown a promising future of applying FIC on SA to further improve the compression performance.

Extracting key postures using radon transform

Abstract: Human key posture extraction will benefit for human action recognition, human action retrieval, human behaviour understanding and so on. This paper proposes an approach to select key postures from a human action video based on Radon transform. Cluster is used on the Radon transform to select the final key postures of human action video. The approach does not require motion extraction from the human action video. The experiments results show that the proposed approach is efficient.

Bi-cubic Interpolation for Image Conversion from Virtual Hexagonal to Square Structure.

Abstract: Hexagonal image structure represents an image as a collection of hexagonal pixels rather than square pixels in the traditional image structure. However, all the existing hardware for capturing image and for displaying image are produced based on square pixel image structure. Therefore, it becomes important to find a proper software approach to mimic the hexagonal structure so that images represented on the traditional square structure can be smoothly converted from or to the images on the hexagonal structure. For accurate image processing, it is critical to best maintain the image resolution during the image conversion. In this paper, we present an algorithm for bi-cubic interpolation of pixel values on a hexagonal structure when convert from the hexagonal structure to the square structure. We will compare with the results obtained through bi-linear interpolation for the conversion. Our experimental results show that the bi-cubic interpolation outperforms the bi-linear interpolation for most of testing images at the cost of slower and more complex computation.

An efficient algorithm for image retrieval through fusion of two clustering approaches based on combined features

Abstract: This paper proposes new approaches for image representation to bridge the gap between visual features and semantics. Two new combined feature extraction approaches are used to extract significant features from images. Each approach is a hybrid of two feature extraction methods and tries to capture both colour and texture information. In order to improve the query processing time and avoid the linear search problem, a clustering technique is applied on the image dataset according to each feature extraction approach. The clustering outcomes of the two feature extraction approaches are combined together using a decision fusion technique. The fused results show an improvement over any single approach. An implemented prototype system demonstrates a promising retrieval performance examined on 1000 colour images from CORAL dataset in comparison with a peer system in literature.

Subspace Analysis Methods plus Motion History Image for Human Action Recognition

Abstract: This paper proposes a new human action recognition method which deals with recognition task in a quite different way when compared with traditional methods which use sequence matching scheme. Our method compresses a sequence of an action into a Motion History Image (MHI) on which low-dimensional features are extracted using subspace analysis methods. Unlike other methods which use a sequence consisting of several frames for recognition, our method uses only a MHI per action sequence for recognition. Obviously, our method avoids the complexity as well as the large computation in sequence matching based methods. Encouraging experimental results on a widely used database demonstrate the effectiveness of the proposed method.

Pedestrian detection using hybrid statistical feature

Abstract: A novel approach for walking people detection is proposed in this paper, which is inspired by the idea of gait energy image (GEI). Unlike most of common human detection methods where usually a trained detector scans a single image and then generates a detection result, the proposed method detects people on a sequence of silhouettes which contain both appearance characteristics and motion characteristics. Thus, our method is more robust. Encouraging experimental results are obtained based on CASIA gait database and the additional non-human objects data.

A comparison of reduced support vector machines

Abstract: In this paper, we first briefly review some knowledge of Support Vector Machines (SVMs). This includes not only SVM's dual forms and solution forms but also the general process and properties of SVM. Then aiming at speeding up SVMs, some kinds of Reduced SVMs (RSVMs) are discussed in detailed. A comparison among them is presented in several aspects. Finally, we show research issues that need to be resolved or investigated further. A number of future trends are also briefly sketched.