Image Recommendation System Based On The Image Features Obtained From Deep Learning

With the rapid development of information technology and graphics technology,information of image is in fast-growing. In the field of scientific research and business applications also increasingly strong demand of image content information. It is difficult for users to quickly find out the image of interest from the vast amounts of image library, so the research direction of image recommendation appears. Deep learning is a new research direction in the field of computer science and machine learning. Its ultimate goal is to make the machine have analysis ability, the machine would be able to identify data such as text, image and sound just like human beings do.In this thesis, deep learning method is combined with image retrieval for research.It makes the results only related to the content of the image itself, reduces the affects of image recommendation’s human’s subjective factors, shooting environment and the influence of the filming equipment.This thesis researches deep learning related development course and the latest research results. Analyse the related technology used by the current image recommendation system and the existing problems. On the basis of the characteristics of deep learning of multiple nonlinear transformation, as well as to the image of multi-layer abstract ability, in the theory of deep learning Convolutional Neural Network is applied to e-commerce image recognition and e-commerce image recommendation. Studied the Convolutional Neural Network is applied to the search performance in image recommendation system. In this thesis, the main work is as follows:1. The image data acquisition with labelsResearched the implementation scheme of the crawler. We can through the crawler technology to domestic mainstream e-commerce site commodity image and the related web page text. Research and solution the problem about web page incremental Crawl ing and removing duplicated URL, unstructured web text entity extraction and structured storage and index, colleagues based on web text information of the image tag.2. The use of the deep learning method in image recommendationThe information of e-commerce image are varied and complicated with complex background, besides,the content of searched images is influenced by the environment and equipment for filming, so this article use the method of deep learning technology for image feature extraction, depth of learning through multiple layers of nonlinear transform of image feature. This method through the study of the depth of the training model to simulate the human visual system understand the process of image content, the image characteristics in closer to the abstraction of human visual perception characteristics, not only can effectively prevent the interference of background of image, but also can resist various deformation, scale transformation, color transformation, can effectively improve the search accuracy when the input image without a label.3. Large-scale image index and retrieval recommendationUsing local sensitive hash algorithm combined with a distributed system, we set up the image index and distributed retrieval platform. In addition, combined with the imageretrieval technology based on deep learning, we reduced the subjective influence of artificial labeled images greatly and optimize the final effect of the image retrieval.4. The retrieval result evaluationThis dissertation used Normalized Discounted Calculate Gain values to evaluate the retrieval results, Normalized Discounted Calculate Gain values comprehensively considered the similarity of retrieval result and target image, it also considered the array location of the similar or same retrieval results.