Image Retrieval And Classification Based On Multi-example Learning

With the rapid development of Internet technology and artificial intelligence,the management of picture information has become a problem that cannot be ignored.Multi-example learning,a new framework of machine learning,has attracted much attention since it came into being.In this paper,the basic principle of multi-example learning is studied,and based on this,three aspects of salient target detection,image classification and retrieval are explored.Firstly,this paper designs a saliency target detection algorithm based on random forest for multi-example learning to solve the weak correlation problem between multi-example learning examples.The target sampling was taken as a multi-example package,and the superpixel in the package was taken as an example,and the salient target detection problem was transformed into a multi-example learning problem by using the randomly selected high-quality features.In this method,the high-level target information of the scheme and the intermediate cue of the superpixel are considered comprehensively,and the feature association of multi-example salient target detection is significantly improved.In this method,the edge of the prominent target is enhanced,the prominent target area is more obvious,and the background area is more effectively suppressed.Secondly,aiming at the ambiguity of multi-semantic image classification and multi-object classification,this paper designs an image classification algorithm based on multi-example learning.Multi-example multi-label learning model is studied to solve the problem of multi-semantic image classification.The image to be processed is taken as the multi-example package,and the segmented part of the image is taken as the example.The multi-semantic image classification problem is transformed into the multi-example multi-label learning processing.Then the multi-semantic image classification problem is transformed into the traditional single label classification problem by using the visual projection function.This method takes into account both the multi-objective problem of multi-semantic image and the efficiency of image classification and achieves good experimental results.Finally,aiming at the complexity of multi-semantic retrieval,a retrieval algorithm based on image hierarchy classification is designed.Firstly,this paper studies the hierarchical classification model of multi-semantic images.According to the search keywords entered by users,the branches related to the keywords are searched and displayed in the hierarchical classification database.Users can select the keywords they are interested in to find the interested images.This method makes the retrieval process of the user more simple and clear,and can obtain more accurate image information according to the prompts.