Image Retrieval Algorithm Research Based On Category Pattern

With the rapid development of computer vision technology,the requirements for image understanding technology are also increasing.Image retrieval as a branch of image understanding direction has very important research significance.This paper studies the image retrieval by forming a more effective image category information for images,aims to train a more effective image retrieval model and improve the image retrieval ability.First of all,an image retrieval model based on the "point-to-face" category pattern is established.On the basis of one-to-one image retrieval,category information is added,image retrieval on a single image is converted into image retrieval on a category image,and the category pattern of the query image is mined,which enriches the semantic information of the query image.At the same time,the rationality evaluation is carried out on two databases with different sizes.The experiment proves that the image retrieval based on the category pattern has better effect than the traditional “point-to-point” method and the hash method.Secondly,an image retrieval algorithm based on object semantic pattern mining is designed,which extends from a single image to its object semantic category.The object semantic category pattern is used to represent the retrieval image and image semantic retrieval.Not only the pattern of the image is mined,but also further highlights the characteristics of the object,and reduces the influence of similar background.Experiments show that the object semantic pattern mining methods achieve better results in image retrieval.Finally,an image retrieval algorithm model based on object category segmentation is constructed.The object detection method is used to segment the object and the background in the image,and the semantic features of the object image are extracted,highlight the category information of the image.That is,the retrieval is performed using the object category information in the image.At the same time,both guaranteed semantic similarity and visual similarity.And finally,the effectiveness of the algorithm is verified on a Pascal VOC 2007 multi-object database.