Inter-intra frame segmentation using colour and motion for region of interest coding of video

In this paper, we present a segmentation method that is based on combined colour with temporal features like motion vectors. In the past segmentation methods for region of interest (ROI) coding has only based segmentation on one feature such as motion, colour and luminance etc [1–2]. The segmentation system that we proposed is based on two features there by combining the strengths of each separate segmentation technique e.g. fidelity and consistency of region combined with moving regions. First of all we segment image (pixels level) inside frame based on colour and then in the second step we segments the images (block level) in a set of frames based on motion. This is initial work of our project where we want to compress video by removing perceptual redundancy based on human visual perceptual capabilities. Human perceive images in term of objects which can never be captured as a whole image or its global features. In order to identify the Region of Interest (ROI) from the frames (and sequence of frames), this efficient segmentation is very important. As a first step, for colour based image segmentation we use an efficient method where speed is our primary concern because we have to process multi frames per second and for motion based image segmentation we use block based motion estimation combined with a motion threshold. Quality of the motion segmentation will be considered in the second step where we use morphological operators to remove interior and exterior noise. Since segmentation is based on blocks, where we have quite less number of pixels than in the image, so speed in not such a big issue. For the third step we explore how to combine the motion and colour segmented images. Our first algorithm is based on combining the edges of the colour segmented image with the edges of the block based motion estimated image. We are working on other algorithms to try and obtain more concise and consistent results. This paper describes our first approach.