Segmentation And Recognition Of Images And Videos Based On Deep Learning

In recent years,multimedia video has played an important role in information dissemination as an important data carrier.However,the expansion of information has brought great challenges to the accurate retrieval of multimedia video.The earliest multimedia video retrieval technology was completed by manual labeling.As the number of videos has increased dramatically,manual labeling has become an impossible task.Later,a retrieval system based on low-level features was established,but the effect was not good.In recent years,video segmentation has become a hotspot in video retrieval research.Video segmentation can separate meaningful entities from video sequences and improve the accuracy and efficiency of retrieval.With the continuous development of deep learning,deep learning technology has made great progress in computer vision tasks.At present,the methods of image segmentation and recognition are based on deep learning.Through deep learning to learn high-level semantic features,image segmentation and recognition can be accurately performed.With the great success of deep learning in image segmentation and recognition,everyone began to use deep learning to segment and identify video.This thesis proposes a method of segmentation and recognition of image and video based on deep learning.Image segmentation and recognition is based on the instance segmentation method,based on the Mask R-CNN network,an example segmentation network is given,by redesigning the Mask R-CNN mask branch structure,the instance segmentation is improved and accelerated.More accurate boundary information is obtained by increasing the resolution of the ROIAlign layer on the mask branch and using the method of feature fusion before and after.Using depth separable convolution without reducing the accuracy of the algorithm reduces the training parameters and improves the efficiency of the algorithm.The segmentation and recognition of video is based on image segmentation and recognition,the steps include lens segmentation,key frame extraction and segmentation and recognition.In this thesis,the segmentation method adopts the method based on ?2 histogram,and only needs to consider the pixel distribution,which can detect the lens with larger change.Key frame extraction uses a content-based analysis method that adaptively selects keyframes as the lens content changes.The extracted keyframes are segmented and identified using an example segmentation network model.Experiments show that the proposed algorithm can effectively segment and identify image video.