Font Size: a A A

Researches On Svm Model For Segmentation Of Color Image Based On Visual Attention

Posted on:2013-05-05Degree:MasterType:Thesis
Country:ChinaCandidate:W T GuoFull Text:PDF
GTID:2248330374456534Subject:Systems Engineering
Abstract/Summary:PDF Full Text Request
How to separate the target from the background of image is a problem in the Machine Vision field, and image segmentation is the key to solving this problem. Image segmentation is to separate the different regions that has special meaning of the image, and the segmentation results will affect the process of image understanding and analysis directly. With the more widespread use of the color image, color image more in line with the human visual habits, so more and more attention for the study of color image segmentation method. Image segmentation method based on Support Vector Machine (SVM) is as a classification problem, and this method requires artificial selection and labeled training samples usually, thus the training samples selected by subjective factors will not only reduce the self-adaptive image segmentation, and the quality of the training samples selected will directly affect the classification performance of SVM. Because the image processing method based on visual attention take into account the characteristics of human visual and its processing results more consistent with human visual habits, more attention has been paid. Usually, salient region detection method based on visual attention get a salient regional boundaries are not accurate, and it is difficult to use for image segmentation. This paper takes full advantage of SVM’s learning and classification capabilities, and combine with detection methods based on the human visual habits, can solve the above problems.The content of this paper are summarized as follows:(1) We proposed an SVM segmentation method based on visual attention for color images of natural scenes, known as VA-SVM segmentation method, and research the new method preliminary. The new method pre-segments image based on visual attention, and will automatically select and label SVM training samples on the basis of pre-segmentation, then using the SVM classifier trained on the whole image.(2) We found VA-SVM segmentation method exist some problems and made some improvements to the VA-SVM segmentation method, and proposed VA-SVM segmentation method based on the quantified color space and VA-SVM segmentation method combined with the spatial information. Speed and performance of the algorithm has been further improved.(3) Experiments were carried out on a large number of images of the image database and the Internet to VA-SVM segmentation method and the improved segmentation method respectively. The experiments show that the results of segmentation are not only good but also consistent with human visual habits through comparing with manual labeled results in the image data base. The experiments also show that the problem of the VA-SVM segmentation method has been resolved.Image segmentation is more important and more popular research direction in machine vision, and combined SVM with good learning and classification performance and image segmentation method based on visual attention more in line with the human visual habits for image segmentation can achieve better results, and makes the results more humane. This paper can not only work done by the rich SVM theory and research, but also to broaden the scope of application of the visual attention mechanism and SVM, and has important theoretical significance and application value.
Keywords/Search Tags:Visual attention, Saliency Detection, Support Vector Machine, Image segmentation
PDF Full Text Request
Related items