Reserch On Interactive Image Segmentaion Relevant Technology

Image segmentation serves as a fundamental element in computer visiontechnology. Image segmentation is significant in digital image processing domain as thetechnique is widely and effectively applied in artificial intelligence field includingimage recognition, image retrieval and image analysis. Due to complexity of colorfeature and texture features encompassed in natural images, satisfying outcome can behardly achieved by fully automatic image segmentation. Under this context, interactiveimage segmentation is introduced, which can precisely separate the targeting objectfrom image through interactive information algorithm the user provided. However, theenormous expense in both time and money incurred during large image processingunder contemporary interactive image segmentation algorithm severely undermining theuser’s working efficiency as well as the utility of image segmentation technology.The purpose of this paper is to discuss how to improve the performance of currentalgorithm. One possible approach is by combining superpixels algorithm, traditionalGrab Cut algorithm and region merging algorithm, which comes up with the idea ofimproved grab cut algorithm and maximal similarity based region merging algorithm.The issues will be discussed in following manner:1. Introducing fundamental theories related to current interactive informationimage segmentation technology and mainly focusing on comparing theperformance and shortcomings of traditional graph cuts algorithm and grab cutalgorithm.2. Revealing7common superpixels algorithms. The algorithms will be tested andcompared on their quality and performance. Based on the findings, one SLICalgorithm in accordance with our demand will be selected.3. Presenting improved Grab Cut algorithm which is the combination of SLICalgorithm and Grab Cut algorithm. Based on the comparison of test results,improved algorithm is at least4times faster than traditional Grab Cut algorithmin terms of running speed. 4. Improving traditional region merging algorithm by adding regional maximalsimilarity rules so that segmentation won’t require manual setting of threshold.Besides, the watershed algorithm in initial segmentation phase is replaced bySLIC algorithm. Our region merging algorithm take less time and can segmentmulti-object in one image.