| Summary: | 碩士 === 義守大學 === 資訊工程學系碩士班 === 93 === Recently, images are used extensively to represent the information in computer technology. The use of low-level features to retrieve relevant information from image and video database much attention in recent years. In this thesis, we will develop a technique to detect the salient region of the images, and the retrieval performance can be efficiently improved.
Many image segmentation techniques have been developed in the literatures. Generally, these techniques are based on region and/or edge information. However, these methods can not achieve satisfactory performance so far.
In this thesis , we present a technique to segment the salient region in an image, and wish to address the drawbacks of conventional segmentation methods. We analyze colors histogram by the method of non-parametric density estimation, and then obtain a reasonable quantization of image. We eliminate the small region using a simple clustering rule. Next we will define the properties of salience region. Accordingly, we will develop an algorithm to achieve salient region segmentation. Experimental results show that the new technique can efficiently segment the salience region the images.
|
|---|
