Based On Wavelet And Quaternion Fractional Fourier Transform For Visual Saliency Detection

With the development of computer technology and related image processing equipments, color image has become the mainstream of all kinds of multimedia information. So color image processing is becoming an urgent demand of people, and the range of application is also becoming more and more widely. Color image contains abundant visual information, and there exists a close spectrum relation among three color channels, so the methods for gray image processing are difficult to apply directly to the color image processing. Meanwhile compared to gray image processing, color image processing is more difficult. How to better use the abundant visual information of the color image from the perspective of human vision is the key to color image processing. Therefore, the processing of salient object detection is studied in this thesis, based on deniosing firstly.Firstly, it describes the research status of color image processing and main contents in this thesis, and then from the point of human vision, it describes basics of color optical and classification of color space model, including conversion relations. Finally, it describes the methods of denoising and salient object detection for color image respectively in details.For the impulse noise which may be introduced in color image, a switching vector-based filtering method for color image is proposed for pre-processing in the third section of this thesis, in order to prepare for subsequent processing. Firstly, it introduces the common noise models, features, and the traditional filtering methods of color image. Finally, a new vector-based filtering method for color image is proposed based on MCWVMF algorithm, introducing noise detection mechanism and conbining with the human visual perception to construct the weight factor, which has a good filtering effect.Processing of color image from the perspective of human visual perception has become a focus in computer vision and artificial intelligence processing field, this thesis puts forward an algorithm of salient object detection for color image in the fourth chapter. Firstly, it introduces the physical structure of the human eye and several common models of human visual perception. Finally, combined with the quaternion theory, a new algorithm for salient object detection is proposed, which takes full advantage of the characteristics of wavelet multi-scale and multi-resolution, the phase spectrum information of Quaternion Fractional Fourier Transform. So it can detect the salient object in color image effectively.