Image segmentation using human visual system properties with applications in image compression
Abstract
In order to represent a digital image, a very large number of bits is required. For example, a 512 x 512 pixel, 256 gray level image requires over two million bits. This large number of bits is a substantial drawback when it is necessary to store or transmit a digital image. Image compression, often referred to as image coding, attempts to reduce the number of bits used to represent an image, while keeping the degradation in the decoded image to a minimum. One approach to image compression is segmentation-based image compression. The image to be compressed is segmented, i.e. the pixels in the image are divided into mutually exclusive spatial regions based on some criteria. Once the image has been segmented, information is extracted describing the shapes and interiors of the image segments. Compression is achieved by efficiently representing the image segments. In this thesis we propose an image segmentation technique which is based on centroid-linkage region growing, and takes advantage of human visual system (HVS) properties. We systematically determine through subjective experiments the parameters for our segmentation algorithm which produce the most visually pleasing segmented images, and demonstrate the effectiveness of our method. We also propose a method for the quantization of segmented images based on HVS contrast sensitivity, and investigate the effect of quantization on segmented images. We apply these segmentation and quantization methods in a new compression technique which fits into the category commonly known as "second generation" image compression methods. Our compression method is designed for application single-frame images (i.e. not time-varying imagery). Other segmentation-based image compression techniques have typically represented the image segments by encoding the boundaries of the segments. We propose the use of morphological skeletons to represent the segments. The morphological skeleton of an image is similar to the medial axis. We describe the application of mathematical morphology to generate skeletons for the image segments, and discuss the advantages and disadvantages of using morphological skeletons in segmentation-based image compression.
Degree
Ph.D.
Advisors
Delp, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.