Group Image Coding Algorithm Based On Object And Quadratic Fitted Photometric Transformation

In recent years,many large-scale technology companies have provided cloud storage services because of the emergence of cloud storage.Many users choose to save image data by using cloud storage.And a large amount of image data will occur in the cloud.How to efficiently compress and store image data of the cloud is the key of research.Traditional approaches of image compression coding such as JPEG,MPEG,etc.use a single image to compress and store,and the similarity among images is not taken into consideration so that the coding efficiency for the entire image set is not high enough.Therefore,in order to use the redundant information between images effectively and further improve the compression efficiency of cloud image sets,a joint compression coding of image sets,ie,group image coding,is proposed.The existing frame coding framework and principles is introduced in this thesis.Then,under the existing framework,this paper focuses on three aspects and conducts simulation experiments.Firstly,a photometric transformation based on quadratic fitting for the current photometric transformation is proposed,which makes the result of photometric transformation closer to the target image and improves the coding efficiency of the group image.Then,an area-based image distance description method is implemented.The algorithm uses triangulation to calculate the coverage area of the SIFT matching points.Then the area and the distance of the SIFT matching point is calculated according to the relevant formula as the area-based image distance.The area-based image distance description method more precisely describes the similar information between images,generates a more efficient coding structure,and improves the final coding efficiency.Finally,a multi-object coding method based on the single reference coding in the existing group images is proposed in this thesis.This method uses object detection technology to identify the objects contained in the image,find appropriate reference images for the objects in the image and the unrecognized remaining parts,and then performs corresponding multi-reference prediction image generation.Finally,block-based motion compensation coding is performed.The end result significantly improves coding efficiency.It can be seen from the experimental results that,for image sets with large differences in luminosity,greatly improves the coding efficiency is greatly improved by the Quadratic Fitted Photometric Transformation.Under the same bit rate,the Quadratic Fitted Photometric Transformation of the PSNR improves the 0.12 d B.Using the area-based image distance description method to test the coding efficiency of the image set compared with the original SIFT matching point,the PSNR is improved by about 0.1 d B at the same bit rate,which improves the coding efficiency of the group image.Comparing the object-based multi-reference group image coding and the current single-reference group image coding,the experimental results show that the PSNR of the multi-reference group image coding based on the object increases by about 0.1d B to 0.6d B at the same bit rate.