Wei-Han Chang

A fast MPEG-7 dominant color extraction with new similarity measure for image retrieval

Dominant color descriptor (DCD) is one of the color descriptors proposed by MPEG-7 that has been extensively used for image retrieval. Among the color descriptors, DCD describes the salient color distributions in an image or a region of interest. DCD provides an effective, compact, and intuitive representation of colors presented in an image. In this paper, we will develop an efficient scheme for dominant color extraction. This approach significantly improves the efficiency of computation for dominant color extraction. In addition, we propose a modification for the MPEG-7 dissimilarity measure, which effectively improves the accuracy of perceptive similarity. Experimental results show that the proposed method achieves performance improvement not only in saving features extraction cost but also perceptually similar image retrieval.

Broadcast Video Logo Detection and Removing

Video logos can be considered as a form of visible watermark which is a trademark or symbol to declare the video copyright. However, in some cases, it will cause visual discomfort due to the presence of multiple logos overlapped directly or blended with another logo in videos. In this paper, we use an efficient gradient- based average approach to detect the logo firstly; then, the selected logo mask in the entire video sequence will be inpainted by the spatial-temporal image restoration technique. Experimental results show that our method is capable of recovering various kinds of video logos in the visually plausible way.

A template-based baseball video scene classification using efficient playfield segmentation

In this paper, we present an effective and efficient framework for baseball video scene classification. The results of scene classification can be able to provide the ground for baseball video abstraction and high-level event extraction. In general, most conventional approaches are shot-based, which shot change detection and key-frame extraction are necessary prerequisite procedures. On the contrary, we propose a frame-based approach. In our scene classification framework, an efficient playfield segmentation technique is proposed, and then the reduced field maps are utilized as scene templates. Because the shot change detection and the key-frame extraction are not required in proposed method, the new framework is very simple and efficient. The experimental results have demonstrated that the effectiveness of our proposed framework for baseball videos scene classification, and it can be easily extended the template-based approach to other kinds of sports videos.

Image Recovery Based on Effective Image Completion

Exemplar-based inpainting algorithm is a model used for large region removal. It combines the advantages of both inpainting and texture synthesis. This algorithm usually requires large amounts of computational cost for priority calculation and matching iteratively. However, for relatively flat damaged areas, the simple interpolation can work well for image restoration. Inspired by the advantages of the color interpolation and the exemplar-based inpainting methods, in this paper, we propose an adaptive restored approach based on gradient-based analysis. The evaluation shows that our method can avoid blurring artifacts and preserving correct propagated structures in the faster way.

Template-Based Scene Classification for Baseball Videos Using Efficient Playfield Segmentation

A robust scene-classification algorithm is able to provide the ground truth for video abstraction and high- level events extraction. In this paper, an efficient playfield segmentation using learning Vector Quantization (IVQ) is introduced, which is able to adapt to the variations of field colors in diverse baseball videos, and then we propose a reduced filed map feature that possesses field-class concept rather than low-level feature and it can also accelerate retrieval performance. Finally, a template-based learning algorithm is proposed for scene classification without shot detection or keyframe extraction in advance. Experiments with the inside and outside tests show that our method is capable of classifying various scenes reliably.

AD efficient synthesis approach for perspective distorted textures

In image-based rendering (IBR), it is commonly employed textures which extracted from reality onto virtual worlds to synthesize novel views, and the patch-based sampling has proven superior in complexity and synthesis quality. However, the conventional synthesis approaches can not be applied directly to the texture which is captured by camera with perspective-like projection, because such textures content have unequal object scale from near to far (perspective distorted textures). In this paper, we develop an automatic and effective approach to synthesize the structured textures with perspective distorted property. The idea behind the work is that the structured textures generally have significant line composition. Therefore, we extract all possible variations from distorted texture as parameterized shapes in binary image, and the Hough transform is applied to detect all possible line structures, and the set of main control points can be identified through Hough transform for planar reconstruction. Experimental results indicated that the proposed texture synthesis algorithm can achieve surprising visual quality for various perspective distorted textures.

An Efficient Baseball Playfield Segmentation Based on Learning Vector Quantization

The segmentation of playfield is essential because it can offer higher level content analysis for sport videos. In this paper, a simple but efficient classification scheme is introduced which is able to adapt to the variations of field colors in diverse baseball videos. First, we utilize learning vector quantization (LVQ) to classify the grass and soil colors of playfields in YUV color space, and then propose the filed map feature that possesses class concept rather than low-level feature and it can also preserve the layout of playfield. Experimental results using three different popular baseball video types revealed that the proposed method is robust and can recognize grass soil and other samples accurately.