Faliang Chang

Automatic License-Plate Location and Recognition Based on Feature Salience

License-plate recognition plays an important role in numerous applications, and a number of techniques have been proposed. In this paper, a novel method to recognize license plates is presented. First, the license plates are located using salient features. Then, each of the seven characters in a license plate is segmented. Finally, the character recognizer extracts some salient features of the characters and uses a feature-salience classifier to achieve robust recognition results. In the experiments, 1176 images that were taken from various scenes and conditions were used, and 32 images out of the 1176 images failed to correctly locate the license plates, which amounts to a success rate of 97.3%. In the experiments on identifying license characters, we used 1144 images, for which license plates have successfully been located and out of which 49 images failed to identify the characters; the rate of successful identification is 95.7%. Combining the preceding two rates, the overall rate of success of the developed method is 93.1%.

Automatic facial expression recognition using local binary pattern

In this paper, an efficient method for human facial expression recognition is presented based on LBP descriptor. Firstly, the face normalization is employed in order to improve the efficiencies of extracting facial features. Then, the feature blocks which mainly contain the eyes and mouth were obtained, and each feature block image was divided into several overlapping sub-regions equally to extract LBP histograms. After that, the PCA method is used to learn the structure of the expression in the LBP feature space. Finally, the recognition experiment is conducted on the JAFFE facial expression database by using the nearest neighbor classifier. Experimental result shows competitive performance.

PTZ camera target tracking in large complex scenes

This paper presents a real-time and automatic video tracking system with a single pan-tilt-zoom (PTZ) camera. Compared with fixed camera, it can enlarge the surveillance area. By using three-dimensional background-weighted histogram in HSV color space, an improved mean-shift algorithm is proposed, and the algorithm can track target with multiple colors in real time. To keep the target in the center area of the cameras FOV, the information acquired from the tracking module is used to control PTZ camera actively. After the movement of the PTZ camera, the target is still in the center area of the cameras FOV. To realize the tracking to the same target, we employ the template matching method to find out the target. Based on motion history image (MHI), which provides the moving blob, a new tracking method is adopted to solve the strong maneuvering target tracking, partial occlusion or the erroneous template matching problem. Experimental results indicate that the system can perform well in large complex scenes.

Infrared small target detection algorithm based on feature salience

It is an important and challenging problem to detect small targets in cluttered scenes with low signal noise ratio (SNR) in infrared (IR) images. In order to solve this problem, a method based on feature salience is proposed for automatic target detection against a complex background. First, in this article, the system utilises the average absolute difference maximum (AADM) as the dissimilarity measurement between targets and the background region to enhance targets. Second, the minimum probability of error has been used to build the model of feature salience. Finally, by calculating the probability of features, this method solves the problem of multi-feather fusion. Experimental results show that the algorithm proposed has better performance with respect to probability of detection. It is an effective IR small target detection algorithm against complex backgrounds.

The hausdorff distance template matching algorithm based on kalman filter for target tracking

The Hausdorff distance template matching algorithm is one of the most popular algorithms recently for its robustness to the noise and the occlusion. But if the binary image that has lots of edge pixel points, this algorithm needs a long time to calculate the distance one by one, so we propose the small region filtration method to reduce the computing time. The experiment results show that the algorithm reduces the time greatly. In addition, because the size of template is very small in contrast to the size of the image, we use search windows in the tracking process, and the search windows are predicted by the Kalman filter.

A novel algorithm of license plates automatic location based on texture feature

Locating the vehicle license plate plays an important role in the vehicle license plate automatic recognition system. A novel locating approach based on the texture feature is presented in this paper. Analyzing the texture of plate image, this paper makes one dimension cycle clear method for map of horizontal projection. This method can effectively filter the noise and clutter around car license plates, and accurately locate them. A set of experiments has been performed to prove the robustness and accuracy of the approach. For many images collected from a large traffic crossing, the experimental results show that 98.5% of them are correctly segmented.

Memory-Based Multi-camera Handover with Non-Overlapping Fields of View

Object tracking is an important task within the field of computer vision, and the multi-camera tracking with disjoint view is more applicable. This paper focus on introducing human memory mechanism into multi-camera human tracking problem that the Field Of View (FOV) of cameras are not necessarily overlapping, and proposing a camera handover scheme based on memory. In the modeling process, every target goes through transmission and storage of three spaces: sensory memory, short-term memory and long-term memory. After learning, memory-based target handoff can remember target appeared earlier, and when faced with similar goals, it can extract and activate the target in memory in time, so that it can quickly achieve object handoff between different camera tracking. Preliminary experiments show that this scheme is effect in camera handover and multi-camera human tracking.

A IHS-WT remote sensing image fusion method based on dynamic weighting of regional multi-features

This paper proposes a new IHS-WT method for remote sensing image fusion method with dynamic weighting of regional multi-features, based on the analysis of the advantages and disadvantages of classical IHS and wavelet transform fusion methods. It combines the advantages of the IHS transform and wavelet transform to achieve a better fusion result, the new I component can be obtained by fusing the wavelet coefficient data of the histogram-matched panchromatic image and the I component through adaptive weights based on window region features. Experimental results indicate that the proposed method has apparent advantage in reservation of spectral information and spatial details enhancement than other methods.

Targets detection under nature background based on feature salience

It is a difficult problem that targets are detected in nature background, especially in images corrupted by noise. In order to solve this problem, a method based on feature salience is proposed for targets automatic detection in nature background. Firstly, an adaptive wavelet thresholding method is used to denoise for noisy target images. Then minimum probability of error was been used to build to the model of feature salience. By calculating the target feature salience and confidence, this method solves the problem of target multi-features fusion. Experimental results show that it is a robust targets detection method.

A fusion method of metallurgical images based on curvelet transform

On the basis of analyzing several common algorithms of image fusion, a new multi-focus image fusion method based on curvelet transform is proposed according to the question of metallurgical image fusion. Firstly, two different focal images were decomposed using curvelet transform respectively, and then in the curvelet domain of the two transformed images, the new curvelet coefficients were acquired by adopting a simple fusion rule, which was that the low-frequency coefficients were integrated using the weighted average, and high-frequency coefficients were integrated using choose max. Finally, the fused coefficients are reconstructed to obtain fusion results. Experimental results show that the method is more suitable for metallurgical image fusion than other ways, the fused image will have more information and the observation with naked eyes will be clearer.