Research And Verification Of Image Compression Based On Image Colorization

An image has a similar appearance of some objective objects and is an important source of human understanding of the world.Generally,the amount of data required to describe an image can pose a challenge to the corresponding storage or transmission devices.Therefore,it is necessary to compress the image,effectively.Different from the traditional domain-based image compression algorithms based on Discrete Cosine Transform(DCT)or Discrete Wavelet Transform(DWT)such as JPEG,JPEG2000 and so on,the thesis will discuss image compression from the perspective of image colorization.In the process of applying image colorization to the field of image compression,two major challenges are faced.The first is how to select the chrominance prior information about the image effectively.The second is how to restore an image of high fidelity.Therefore,the thesis will focus on the above challenges,respectively.Firstly,in order to guarantee the fidelity to the restored image,the thesis studies the Levin's colorization algorithm.In view of the two deficiencies existing on the Levin's colorization algorithm such as chroma penetration phenomenon and considering the feature's similarity between images' luminance separately,the thesis introduces the algorithms of neighbor propagation and mixed weight respectively for improvement.The experimental results show that the proposed algorithm not only can effectively alleviate the chroma penetration phenomenon,but also can achieve higher evaluation index.Secondlly,according to the basic idea of example-based colorization algorithm,the thesis puts forward a thumbnail-based compression algorithm that uses the thumbnail and the grayscale image to restoring images.At the compression stage,the image is compressed by extracting the thumbnail and the grayscale.At the decompression stage,the image is restored by image segmentation,region matching,pixel matching and color transferring,based on the information stored in the thumbnail and the grayscale.The experimental results show that the proposed thumbnail-based compression algorithm is comparable to the colotization-based compression algorithm.When JPEG is adopted as an auxiliary storage method,the quality of the restored image will be better than JPEG and JPEG2000 compression standards.Finally,the thesis proposes a representative-pixel-based compression algorithm,which iteratively extracts representative pixels by hierarchical clustering,based on the typical colorization-based compression algorithm.At the compression stage,the representative pixels of the image are iteratively extracted by feature clustering,chroma clustering,redundancy elimination and temporary image colorization.The further compression is achieved by run-length encoding.At the decompression stage,the image is restored by using scribble-based colorization algorithm based on the information stored in the representative pixels.The experimental results show that the quality of the restored image using representative-pixel-based compression algorithm is better than colotization-based compression algorithm,JPEG and JPEG2000 compression standards.The thesis focuses on the application of image colorization in the field of image compression,which not only enriches the existing image compression methods,but also provides reference to subsequent research on image compression.At the same time,it has some theoretical research values.