Segmentation through DWT and adaptive morphological closing

Object segmentation is an essential task in computer vision and object recognitions. In this paper, we present an image segmentation technique that extract edge information from wavelet coefficients and uses mathematical morphology to segment the image. We threshold the image to get its binary version and get a high-pass image by the inverse DWT of its high frequency subbands from the wavelet domain. This is followed by an adaptive morphological closing operation that dynamically adjusts the structuring element according to the local orientation of edges. The ensued holes are, subsequently, filled by a morphological fill operation. For comparison, we are relying on the well-established Canny's method and show that, for images with low-textured background, our method performs better.