Spatiotemporal Approach for Tracking Using Rough Entropy and Frame Subtraction B. Uma Shankar and Debarati Chakraborty Machine Intelligence Unit, Indian Statistical Institute, Kolkata 700 108, India uma@isical.ac.in, debarati.earth@gmail.com Abstract. We present here an approach for video image segmentation where spatial segmentation is based on rough sets and granular com- puting and temporal segmentation is done by consecutive frame sub- traction. Then the intersection of the temporal segmentation and spatial segmentation for the same frame is analyzed in RGB feature space. The estimated statistics of the intersecting regions is used for the object re- construction and tracking. Keywords: Segmentation, rough entropy, rough sets, video tracking. 1 Introduction In computer vision, detection and tracking of moving object is very important task. The application of object tracking in video sequences has been studied over the years. In this task there are many types of uncertainties and ambigu- ities which is making this task a difficult problem. Over the years researchers have been trying to improve the accuracy and speed in detection and tracking [5,11,13]. Granular computing is a young, though rapidly expanding, and im- portant area of research. Granular computing is the process of dealing with the information granules which is collection of some points similar in some respect and dealing with which proves to be effective for human cognition. In case of im- age and video processing, proper recognition of each part of the image from the available information is a well known problem where partitioning (segmentation or classification) of data set plays a very important role. Understanding of an image depends on efficient partitioning. So, granular computing can be useful for image processing. According to Butenkov [2] the most important problem re- garding granulation of an image is different kind of input information. Here we propose to use Rough entropy as proposed by Pal et al. [7], incorporating some modification for video image segmentation and object detection for tracking in video image sequences. In the present article we have proposed an approach for video image segmenta- tion based on granular computing and rough sets. The contributions of the article are as follows : (i) We propose a method for detection of granule adaptively. Here we decompose the image into homogeneous granules using quadtree decomposi- tion, which takes into account the spatially connectedness and gray level similar- ity, both. (ii) Then a general form of rough entropy function is defined, which can S.O. Kuznetsov et al. (Eds.): PReMI 2011, LNCS 6744, pp. 193–199, 2011. c  Springer-Verlag Berlin Heidelberg 2011