Research On Image Retrieval Algorithm Based On Block Feature

With the advent of the era of big data,huge amounts of image data are flooding people’s lives.How to quickly retrieve the image that people need in many images have always been a popular issue in the field of computer vision.Although technologies such as image processing and image retrieval under the continuous exploration of researchers have achieved very good results in many fields,local characteristics will be lost during the quantification process when the feature extraction of images,and the global characteristics will also be in quantitative process,and the global characteristics will also be.More difficult reflection of image content information.Therefore,based on the content retrieval of content based on content,this article analyzes how the image is divided into blocks,proposed a new block strategy,and combines it with the convolutional neural network to describe the detail information of the image.In this way,the accuracy of image retrieval is increased.The main work content of this article is as follows:1 In order to improve the ability to describe the image information,a method retrieval method based on K-means blocks is proposed.First,use the K-means algorithm marker image.Then,the obtained sub-images are treated morphologically,and the edge details of the image subject are retained at the same time on the images.Finally,the sub-image after the image is divided into the sub-image,and the similarity is calculated to complete the image retrieval task well.The sub-block obtained by the K-means block method not only contains the spatial information of the image,but also overcomes the shortcomings of the uniform block method to destroy the content of the image target content.It has certain theoretical and application significance.2 In order to solve a single image feature,it cannot fully describe all the information problems of the image,and proposes a image retrieval method based on the K-means block and Mobile Netv3.First of all,in order to increase the weight of the image subject,the segmented sub-images are superimposed with the original image.Secondly,in order to make the Mobile Netv3 network more good at extracting the high-level semantic characteristics of the image,use the ECA attention for deep convolutional neural network.Essence Then,two pooling paths are introduced in the Mobile Netv3 network,which enables the final features to include high-level and low-level semantic characteristics.Finally,use Corel-5K as a standard database for image retrieval experiments,and use the accuracy and retrieval rate as an evaluation indicator for image detection.K-means divided into the Mobile Net V3 method,which effectively improves the accuracy of retrieval.