ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

Superiority Of Graph-Based Visual Saliency GVS Over Other Image Segmentation Methods

Journal: International Journal of Scientific & Technology Research (Vol.6, No. 2)

Publication Date:

Authors : ; ; ;

Page : 14-21

Keywords : computer vision; image segmentation; visual saliency detection; graph-based algorithm.;

Source : Downloadexternal Find it from : Google Scholarexternal

Abstract

Although inherently tedious the segmentation of images and the evaluation of segmented images are critical in computer vision processes. One of the main challenges in image segmentation evaluation arises from the basic conflict between generality and objectivity. For general segmentation purposes the lack of well-defined ground-truth and segmentation accuracy limits the evaluation of specific applications. Subjectivity is the most common method of evaluation of segmentation quality where segmented images are visually compared. This is daunting task however limits the scope of segmentation evaluation to a few predetermined sets of images. As an alternative supervised evaluation compares segmented images against manually-segmented or pre-processed benchmark images. Not only good evaluation methods allow for different comparisons but also for integration with target recognition systems for adaptive selection of appropriate segmentation granularity with improved recognition accuracy. Most of the current segmentation methods still lack satisfactory measures of effectiveness. Thus this study proposed a supervised framework which uses visual saliency detection to quantitatively evaluate image segmentation quality. The new benchmark evaluator uses Graph-based Visual Saliency GVS to compare boundary outputs for manually segmented images. Using the Berkeley Segmentation Database the proposed algorithm was tested against 4 other quantitative evaluation methods Probabilistic Rand Index PRI Variation of Information VOI Global Consistency Error GSE and Boundary Detection Error BDE. Based on the results the GVS approach outperformed any of the other 4 independent standard methods in terms of visual saliency detection of images.

Last modified: 2017-06-11 22:59:11