Research And Implementation Of Content-based Image Retrieval Of Commodity Logo

In the field of Internet e-commerce,it is common for sellers in the same industry to plagiarize and copy information from other stores,and plagiarized images are more difficult to detect similarity than text messages because plagiarists process images to make the processed The image is not easy to detect and is similar to the original image.The efficiency of manual comparison is low and the cost is high,which requires a retrieval system based on an algorithm that can quickly and accurately retrieve images to solve this problem.Image retrieval systems have a wide range of applications,such as information retrieval,trademark and intellectual property protection.Therefore,it is of great significance to study the image retrieval system.After summarizing previous studies,this paper proposes a fast nearest neighbor matching image retrieval algorithm based on SIFT feature extraction and an image retrieval algorithm based on twin convolutional neural networks.Based on the above,the specific work of this article is summarized as follows:(1)The paper first studies the image retrieval algorithm based on SIFT and fast nearest neighbor search,first uses SIFT algorithm to extract image feature points,then uses fast nearest neighbor search to match the image feature points,and finally completes the image features of the two images Quantification of matching.The robustness of the algorithm is verified through experiments,which can well deal with the situations of cutting,rotating,adding filters,increasing contrast,adding watermarks,and finally with color histogram,image hashing and other algorithms in the data set Flickr-On the 27 th,an experimental comparison was made,and the average precision was used as an indicator to verify that the algorithm had higher accuracy than the above two algorithms.(2)The paper then studies the image retrieval algorithm based on the Twin Convolutional Neural Network(SCNN),first using SCNN to extract image features and quantify the similarity between the features.Then through experiments,the generalization ability of the model trained on the Flickr-27 data set is verified.Theaverage precision of each class in the training set and the test set is no more than 8%,indicating that the model has a strong generalization In the end,the experiment is compared with color histogram,image hash,SIFT and other algorithms under the same conditions.The data results show that the accuracy of the algorithm is 40%higher than that of SIFT,which verifies the high performance of the algorithm.(3)The thesis finally designs and develops an image retrieval system based on commodity content.The Flickr-27 data set is preprocessed and imported into the system's database,and the algorithm studied in this article is embedded into the system.Finally,the retrieval effect of the system is tested.The retrieved image is Adidas,and the returned image is 36.Next,the SIFT algorithm has 12 images correctly retrieved,and the SCNN algorithm has 30 images correctly retrieved,further verifying the high performance of the SCNN algorithm.