Fast Group Image Coding Algorithm Based On Reference Image Synthesis

In recent years,people choose to save image and video data to the cloud with the development of cloud computing and cloud storage.Efficient and high-speed compression of image data stored in the cloud has become an important research direction.In order to make full use of the correlation between image among the image set and further improve the efficiency of image compression coding,scholars have proposed the group image coding algorithm.The coding efficiency of the group image coding algorithm while compress the whole image set is higher than that of single image compression.However the algorithm complexity is very high,and the encoding rate is greatly reduced.In this thesis,we first introduces the existing group image coding framework and the general principle,and then proposes a new scheme for the reference image preprocessing process,and puts forward the corresponding improvement scheme for the object detection algorithm in the coding structure determination module,in order to reduce the coding complexity and improve the coding rate while ensuring the compression coding efficiency.Firstly,a group image coding algorithm based on reference image synthesis is proposed for reference images after perspective transformation and luminosity transformation.In this algorithm,the reference images of multiple object images in the same image are synthesized into two reference images,and the relevant information of the object is concentrated in two reference images as far as possible.For the background part of the target image after the object is removed,the corresponding multiple background reference images are synthesized into two reference images with the strongest correlation with the target image,so as to greatly reduce the number of reference frames and greatly improve the encoding rate without losing the compression coding efficiency.Secondly,through further learning of YOLO algorithm,it is found that the accuracy and speed of YOLO-v3 algorithm is significantly improved than that of YOLO-v1 algorithm.Here is a way to improve the experimental means to increase the encoding rate.Through experimental simulation of the improved and realized scheme,the simulation results show that: first,the encoding efficiency of the reference image synthesis algorithm is basically the same as that of the non-use of the algorithm,and the encoding rate is greatly improved.Compared with the same rate,the object reference image synthesis algorithm is adopted to reduce the encoding time of the target image by up to 38%-50%,the background reference image synthesis algorithm by up to 20%-53%,and the object and background reference image synthesis algorithm by up to 38%-70%,that is,the encoding rate is greatly improved under the premise of guaranteeing the encoding efficiency.The more simple the image background is,and the more the image set contains the proportion of identifiable objects,the more obvious the improvement of coding rate will be.Secondly,when YOLOv3 algorithm is introduced to replace YOLO-v1 algorithm as the step of cutting object detection,the overall encoding efficiency is basically the same as that of YOLO-v1 algorithm.The target image encoding time decreases by 45%-50%,which means that the encoding rate is greatly improved while guaranteeing the encoding efficiency.