Design And Implementation Of Automatic Filtering System For Violation Picture Based On Feature Extraction Algorithm

With the rapid growth of an e-commerce enterprise business,the number of images processed by the company's website has doubled.Massive picture information has caused great difficulties for the company's employees in image processing.The duplication and similarity between pictures are getting higher and higher,the organization between pictures is getting more and more chaotic,and the risk of ordering,uploading and reviewing are increased.Some pictures may contain illegal information such as guns,pornography,infringement,etc.If the y are not able to use effective means to filter,it will have a great impact on the company's business.In order to improve the existing image retrieval technology,a new algorithm for automatic recognition of prohibited images a feature recognition based image recognition algorithm is proposed.The algorithm is a hybrid algorithm,including HSV color extraction and HOG shape extraction based on global features.Tamura texture extraction,and local feature extraction based on SIFT algorithm,these four algorithms are combined by serial feature fusion and LDA feature dimension reduction to achieve better image recognition.Applying the algorithm to the banned picture recognition work of the company website,combined with the needs of the image reviewers,design and develop an automatic image filtering system based on the feature extraction algorithm.According to the actual requirements of the system,the system is divided into six modules: picture uploading,prohibited sample database building,prohibited f eature tree building,image feature extraction,image feature comparison,and prohibited image recognition.The use case diagram is used to analyze the use case requirements of these modules.The system is based on the OpenCV open source library architecture to efficiently complete the image recognition and comparison.In the process of extracting image feature points of the system,RANSAC algorithm is carried out on the images with noise points in the images to prevent these noises from affecting the process of image feature points identification and leading to recognition errors.After the feature point extraction is completed,these feature points are collected for K-Means cluster calculation.For the uploaded pictures,the process of identifying prohibited feature points of pictures is officially started.This paper discusses the design process of each functional module of the system by using the sequence diagram,and makes the related design of six major modules,namely,picture uploading,prohibited sample database building,prohibited feature tree building,image feature extraction,image feature comparison and prohibited image recognition.Finally,the implementation effect of the main functional modules displayed in the system interface was tested,and the test results showed that the functional modules could be used normally,and the performance met the design requirements.