Research On Image Retrieval Technology Based On Object Detection

With the development of multimedia technology and the popularization of informatization,image content has become more and more complex.Images with multiple targets,multiple tags,and complex backgrounds are more common in life,which brings more challenges to content-based image retrieval.In the face of this problem,some researchers try to apply target detection algorithms to image retrieval,but there are some defects when simply applying target detection to image retrieval.Aiming at these defects,Aiming at these defects,this paper proposes an image retrieval method based on object detection,which combines high-level semantics and low-level color features.The main research work of this article is as follows:First,the target detection algorithm for multi-target multi label image detection will return a number of uncertain target features,which is not easy to calculate the similarity.This method uses the feature output of target detection algorithm with the characteristics of tagging information to binary code the high-level semantic features of the image.Second,the features extracted by the target detection algorithm cannot fully describe the image information and need to be supplemented with other features.Aiming at the shortcoming that the target detection algorithm only extracts the highlevel semantic information of the target,lacks the low-level information of the target,and cannot distinguish well when detecting the same category.In addition,the target detection algorithm only extracts features from the target area,lacks global information and loses background information outside the target,This method uses local color histogram and color moment to complement.The local color histogram information of the target is extracted by using the target location information determined by the target detection,and the auxiliary calculation is carried out during the similarity calculation of the subsequent retrieval to complete the description of the high-level semantic and low-level color features of the target object.The color moment is used to extract the color distribution features of the global image as a supplement to the target features,which reduces the defect that the objects are similar but the background is completely different.When the similarity sequence is returned,the similarity of the image with approximate color distribution is improved.In this paper,the proposed image retrieval method combining high-level semantics and low-level color features is tested on public data sets.Experimental results show that the average accuracy of the algorithm is improved,and the query results are more consistent with the query intention...