The Research On Photo Management Based On Image Analysis

With the high development of the digital technology and the advent of the Internet, human beings exposed to vast amount of digital images. how to find the needed photos accurately, how to adjust the wrong-orientation photos automatically and how to evaluate the quality of photos properly are some significant problems which needs to be solved. This thesis does research on these problems, and the main work can be introduced as follow:First, this thesis is to take emphases on texture features and color features of images. A feature extraction algorithm based on NSCT is proposed to analyze the texture. Experimental results demonstrate that NSCT can extract texture features effectively. A color feature extraction algorithm based on key region is proposed. This algorithm introduced the location features of the image color and improves the performance of the image retrieval. At the same time, the algorithm sets up a simple system of image retrieval using Visual C++.Secondly, features based on discrete wavelet frame transform coefficients are used to image orientation detection. These features combines the edge direction histogram and using ensemble SVM classifier is proved to get satisfying detection accuracy in the situation of low rejection rate, and the accuracy of the indoor image orientation detection is also much better than that obtained from the method based on color moment.Finally, Based on the traditional evaluation method of image quality, the thesis analyzes some factors that affect the quality of digital image and proposes an comprehensive evaluation method combined blurness, contrast, colorfulness and visual saliency.