Dissertation > Excellent graduate degree dissertation topics show

Based on the topological properties of the visual attention model and its application

Author: FangYu
Tutor: GuXiaoDong
School: Fudan University
Course: Circuits and Systems
Keywords: Attention selection Topological properties Empty filter Saliency map Evaluation Method Image segmentation Region growing
CLC: TP391.41
Type: Master's thesis
Year: 2011
Downloads: 95
Quote: 0
Read: Download Dissertation

Abstract

The vision is the most important kind of all human perception, attention selection is an important characteristic of visual perception. We can easily detect and identify different objects in the image, and video, traditional machine vision, however, it is difficult to do this. Psychological research shows that in the process of human visual perception, the the object global topology perception of First perception, followed by visually perceived brightness, color, sports and other local features, parallel input to the visual neurons synchronization. In visual perception, attention selection mechanism played a key role in helping our attention and to extract the region of interest in the scene and target. Attention selection mechanism is divided into two parts of the bottom-up and top-down attention. In this paper, we study the visual attention mechanism, based on the psychology topological perception theory, applied to the topological properties of the bottom-up saliency detection method, a new objective saliency map evaluation criteria will be significant applied to image segmentation. The main work and contributions of the thesis contains the following aspects: (1) the nature of the topology is applied to the attention selection model which proposed a topology-based perceptual characteristics of the white bottom-up saliency detection method. Topology connectivity of visual information, colors, brightness, movement feature extraction, parallel input quaternion model. After a the hypercomplex Fourier Transform, the phase information of the original image. After its anti-transform to the spatial domain, and then filtered, and finally get the attention of the model significantly FIG. This method takes into account the important role of visual perception in the topological properties, based on a well-known psychological theory. The significant effect of our model can reflect the attention information distribution. (2) a new saliency map evaluation criteria. The standard does not require human intervention to rely completely on an objective evaluation. In which the bottom-up attention selection model, the model results are generally rely on attention saliency map performance. How to effectively evaluate the saliency map quality has always been important issues in this area. The existing evaluation methods almost all labor participation, such evaluation criteria with the inevitably subjective factors, credibility and persuasiveness is not high. Our proposed evaluation method is based on the evaluation of the contribution of the model channel, all without human intervention, is an objective evaluation. Based on the proposed evaluation criteria, it is found that some of the problems that exist in the model, has been improved, adjust the weights of the topology channel, moderately reduce its impact on the final result. The effect of the improved model can more objectively reflect the real distribution of attention. 3 significant attention applied to image segmentation, the automatic segmentation of color images. The vast majority of the existing color image segmentation rely heavily on manual marking, or parameter adjustment. We will extract significant information of the image, the target and the background area manually marked improvements to the saliency map automatically marked, and a combination of mean shift the two segmentation method based on the maximum likelihood of regional growth, a new color image automatic segmentation method. Segmentation, the new method has improved, but the most important breakthrough is that it is a fully automatic color image segmentation method does not require human intervention.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile