Content Based Clothing Image Technology Retrieval

With the development of e-commerce, especially the rise of online shopping, the requirement of image retrieval technology is also becoming more and more urgent. The existing clothing image retrieval methods is mostly based on low-level visual feature, such as texture, color, shape features of image. Those image retrieval method could be use in any area, and there are no clothing intrinsic features for retrieval, this inevitably leads to a unsatisfactory effect and a low efficient system. The main work and innovation of this paper are as follows:(1) Reviewed some proven image retrieval technology, including similarity measurement, retrieval performance evaluation method and comparison of image retrieval system.(2) In order to eliminate background interference of clothing image, we introduce Grab Cut algorithm. But it is sensitive to local noise, time consuming and segmentation edge is not accurate. To address those problem, we employ the multi-scale watershed algorithm to de-noise gradient image; We can both enhance the image edge points and reduce the subsequent processing computation; To reduce the loss of image key features, we employ entropy penalty factor to optimize segmentation energy function.(3) We introduce Itti visual attention model to content based image retrieval system. It has the problems like inadequate feature extraction, complex feature synthesis process and feature incompatible with existing retrieval system. To solve these problems, we proposed to improve the low-level visual features, image segmentation and interesting area of Itti model. By using texture features, the texture roughness characteristics could be reflect; By threshold segment saliency map of color, brightness, orientation and texture, making the segmentation more efficient and accurate. Then, using region growing algorithm of seeds point, combining merging rules of interested region, we can get more accurate saliency map of visual features, all of those significantly improve recall and precision of the retrieval system.(4) We proposed a method based on the clothing intrinsic features for clothing image retrieval, including clothing collar, sleeves and buttoned features, by extracting these features and apply it to the clothing image retrival system, the accuracy of the system is significantly improved.