Research And Application Of The Key Technology For Object Based Image Retrieval

The object based image retrieval method retrieves images by the objects of the image,which is also one of the recent research hotspots.This thesis focuses on object detection and similarity measurement,which are the two key technologies on object based image retrieval.The specific work is as follows:1)A multi-scale fully convolution object detection method based on feature Pyramid is proposed.This method predicts the location and scale of objects in images by using the multi-scale fully convolution network which is based on feature Pyramid,and uses a bounding box refinement method which is based on conditional probability to correct objects' position.The method is tested on the PASCAL VOC dataset.The results show that the method can not only detect the object of category-independent accurately,but also detect small objects.2)A method of image similarity measurement based on nonlinear hash coding is proposed.This method constructs a nonlinear coding module based on deep hash,which is used to learn a group of nonlinear hash functions and code the objects in the image,and then measures the similarity of images based on these codes.The method is tested on Oxford Buildings dataset.The results show that this method can accurately measure the similarity of image and greatly improve the accuracy and recall rate of image retrieval.3)Based on the above methods,this thesis implements a painting oriented content based intelligent image retrieval prototype system.This System can retrieve images accurately by objects of the ancient painting and shows higher retrieval accuracy and quality on painting images of three different styles.