Research On Automatic Segmentation Algorithm Of Garment Image Based On Prior Knowledge

Image segmentation technology, refers to a method for processing a particular area of interest is separated from the background image and the extracted, it is a key technology in the field of digital image processing, which is the basis of image analysis and understanding, in many image segmentation algorithms, GrabCut algorithm is a kind of can effectively from complex background image extracting algorithm for interactive image segmentation of the foreground object. Today, online shopping is becoming more and more popular, so the research of image retrieval system is fast becoming the key areas, and the most important step is image segmentation, has also attracted more attention of scholars. And apparel online purchase accounted for a large proportion in online shopping, online shopping for clothing image in this paper, first carry on the localization to the clothing area, and then the GrabCut algorithm is improved, so as to realize the automatic segmentation of the image of clothing, the main work is as follows:(1) A large number of clothing image through extensive research, summed up a series of their common characteristics are presented, using the characteristic of a kind of face detection and localization of clothing clothing area method based on edge detection. The algorithm firstly face detection in the image with the AdaBoost algorithm, the coarse positioning of two kinds of face and non face images using two different methods of clothing area, combined with the edge information of clothing, an iterative algorithm with local and global methods respectively based on clothing edge to face and non face two image for coarse positioning frame adjustment, the final completion of the precise positioning of the clothing area.(2) Through the GrabCut algorithm in-depth study of the classical algorithm, the segmentation accuracy is very high, but the algorithm with artificial take part in the process, so in the processing of large images is time-consuming, low efficiency. In this paper, according to the need of manual annotation in the GrabCut algorithm "hard split box" part of the improved segmentation method is proposed, a fully automatic image of clothing, namely the use of clothing image automatic precise positioning results require manual annotation of "hard cutting box" instead of the traditional GrabCut algorithm, the final completion of the automatic segmentation of the image of clothing.(3) The positioning of the clothing area in the image and the segmentation results were given, and the evaluation index data can be seen in the clothing area, positioning method in this paper has good effect, high accuracy algorithm not only inherits the traditional GrabCut algorithm also abolished the interaction part of automatic segmentation of clothing image improved, so that it can be applied to the mass of clothing image retrieval system, improve its universality.