Dynamic texture segmentation of video using texture descriptors and optical flow of pixels for automating monitoring in different environments

Abstract

Dynamic texture is the texture which is in motion. Segmentation of Dynamic texture is a challenging task, as the texture can change in shape and direction over time. In this paper segmentation of Dynamic textures into distinct regions has been done. For the segmentation of dynamic texture, three different techniques are combined together to obtain better segmentation. Two local texture descriptors called Local binary pattern and Weber local descriptor are used. These descriptors, when used in spatial domain helps to segment a frame of video into distinct regions based on the histogram of the region. Also using the same texture descriptors in temporal domain, it is possible to obtain the dynamic texture in a given video. In addition to these texture descriptors, optical flow of pixels is used for detecting dynamic texture in a video, as optical flow is the natural method for detection of motion. After computing the three different features from multiple split sections of a group of video frames, individual histograms are obtained for each of the split sections of the video. These histograms each are converted to a single value called the Cumulative. The cumulative so obtained is compared with the threshold and filtered to obtain the dynamic texture. Since the histograms are converted to a single value, the computation of threshold is very easy as the whole set of values of Cumulative falls in two different set of values. The magnitude of motion detected depends on the threshold selected.