Research On Image Retrieval Semantic Enhancement Method Fused With Text Features

With the continuous development of information technology such as digitization,digital image information is growing rapidly in both quantity and scale.How to manage and apply these massive image resources has become one of the many important issues that people are facing.In the field of image search research,the early text-based image query technology has been unable to meet many practical needs.In recent years,content-based image retrieval technology has gradually become a new research hotspot.Usually people judge the similarity of two images not based on the low-level features of the image,but based on the content of the image description.However,it is easier for a computer to understand the low-level features of an image,and it is more difficult to obtain the high-level semantic content description of an image.This is the main difficulty of content-based image retrieval research.This paper aims at image retrieval,and conducts related research on the extraction of image text features and image semantic understanding.The main research works are as follows:1.An image text feature extraction method based on attention mechanism is given.The text content contained in the image is used as a target for research.The Faster-RCNN network is used to detect the text area.At the same time,an attention mechanism is added to the network model to reduce the number of candidate frames through the acquisition of the attention area.On the basis of ensuring the detection accuracy,the detection speed of the text area is accelerated.Then,the detected text area is processed,and multiple single text character pictures are obtained through operations such as connected domain detection and character segmentation,and then the convolutional neural network is used to recognize the character image,and finally the image text content is obtained.2.An image semantic enhancement algorithm based on text features is proposed.The Faster-RCNN network is also used to detect the target of the image,and the color characteristics of the image and the spatial relationship between the detected target areas are extracted.Combined with the text features obtained in the research content 1,the recurrent neural network is used to generate the image semantic description.3.A method of image retrieval semantic enhancement based on text features is given.The semantic description of the image generated above is used to obtain the image semantic features through semantic coding,and extract feature from image database according to this method to obtain image feature database.When querying the image,it extracts the semantic features and performs similarity measurement in the feature database.Finally achieve the purpose of image retrieval.For each algorithm proposed in this paper,in the corresponding chapters,they are tested by experiments.The experimental effects indicate that the prototypes and algorithms given are practical and effective.Compared with the existing methods,the method in this paper can obtain more accurate semantic descriptions,and more accurately complete the image retrieval work that fuses text features.It enables people to better manage and apply massive image data.