A simple hybrid image segmentation method for embedded visual display devices

Abstract

Image segmentation plays a major role in computer vision. It is a fundamental task for feature extraction and pattern matching applications. This paper proposes a simple hybrid Image segmentation method which is mainly based on mathematical morphological operations and filtering techniques. The main aim of the proposed hybrid segmentation method is to segment the foreground object in the given image and mark the segmented region with precision. The purpose of developing this method is to identify a prominent single object based photographs automatically in real time. Also the algorithm must work for worst cases (fog, mist, blur, noise etc). This requirement needs a precise segmentation approach which must be computationally less costly and easy to implement with better quality of segmenting the object as Region of interest. The images are at first subjected to Gaussian filtering to make the image smooth for segmentation. Later, applying Sobel edge detection algorithm to detect the edges properly and then applying morphological operations logically arranged in a novel way for morphological image cleaning purposes. In the final stage, the object of interest is segmented and marked which proves the efficiency of the proposed hybrid image segmentation algorithm. Furthermore, the proposed hybrid image segmentation algorithm is implemented in MATLAB (version 7.0 on an Intel P-4 dual core) and evaluated on 350 jpeg images with satisfactory results. The images sizes which were tested are 384 *