IWSSIP 2010 - 17th International Conference on Systems, Signals and Image Processing 182 Segmentation of Soccer Video Transitions R. G. N. Weber Teigão CELEPAR Curitiba-PR, Brazil beta_n@yahoo.com.br J. Facon PPGIA- PUCPR Curitiba-PR, Brazil facon@ppgia.pucpr.br Abstract— In this paper, a novel approach for soccer video transition detection is proposed. This approach identifies scene cuts based on visual rhythm. A mathematical morphological lexicographic order in HSV space is used to detect scene cuts based on its color composition. Grayscale and binary morphological operators are used for false cut reduction. Experiments accomplished on soccer videos of variable quality show the promising aspects of this approach. Keywords: MPEG Video, Visual rhythm, Scene Transition, Mathematical Morphology, HSV color space, Lexicographic order. I. INTRODUCTION With the increasing use of digital videos, the necessity in providing video search functionalities has transformed digital video indexing and processing in important area. In this context, detecting transitions among shots is an important step for digital video segmentation and analysis. Different kinds of transitions are present in digital videos, i.e., abrupt transitions, cuts, wipes, fades, dissolves, zooms etc... And approaches applied on uncompressed or compressed videos, and based on 2D videos or video-to-2D-image transformations to detect theses transitions are available. For instance, approaches based on dissimilarity measures [14] [4], histogram-based algorithms [11], motion-based algorithms [12], contour-based algorithms [15] or yet production-model based algorithms [6]. The visual rhythm has been presented some years ago as a new way to process videos like images. In [9], [7] one can find a complete definition of visual rhythm. On can say that visual rhythm is a single 2D image created by sub-sampling video content from row or column or yet diagonal pixels of each frame. Although a visual rhythm could only appear as a much summarized representation, most importantly, any kind of video effect is present. The discontinuities of texture and color correspond to a new event while texture and color orientation means camera manipulation and object motion [9]. By transforming a video to a single 2D image, the visual rhythm permits to directly apply various image processing techniques. We decided to use the mathematical morphology which represents one of these powerful image processing techniques. There are few approaches based on morphological tools to analyze and detect video transitions and no one based on color mathematical morphology. One can cite [5] to detect both cuts and gradual transitions or [8] for gradual change detection. This paper presents a morphological approach to detect MPEG standard compressed soccer videos cuts from visual rhythm. No specific knowledge or mathematical modeling about soccer video is required. A color morphology based on HSV lexicographic order is employed to detect cuts in color visual rhythm. Grayscale and binary morphological operators are used to reduce false detection. The rest of the paper is organized as follows. Section II reviews the color mathematical morphology operators and the HSV lexicographic order. Section III formally explains the video cut detection by means of visual rhythm. Experimental results over variable quality soccer videos are discussed in Section IV. II. COLOR MATHEMATICAL MORPHOLOGY Like binary and grayscale mathematical morphology, the color mathematical morphology is based on ordination. But differently, imposing an order on color data is not an easy task. Studies have shown that not only choosing an order but also choosing adequate color space is very important to avoid introducing color distortions. Among the variety of color spaces and orders available in the literature, we have decided to use the lexicographic order onto HSV space proposed by [3]. There are three reasons of this choice: The lexicographical order is a complete order like dictionary ordination - The HSV color space has ability in separating luminance and chrominance information by Hue, Saturation and Value components - The lexicographic order proposed by [3] is based on a metric called Chromaticity Constant that reduces Hue and Saturation components to one value. Chromaticity Constant between two three-component vectors ) , , ( 1 1 1 v s h and ) , , ( 2 2 2 v s h does not use the Value component and is summarized as follows: )) , ( |, (| )] , , ( , ) , , [( 2 1 2 1 2 2 2 1 1 1 h h DistH s s Sup v s h v s h C - = (1) Where