Using Extended Three-valued Increment Sign for a denoising model of high-frequency artifacts in JPEG images by estimation of specific frequency
Reiko Hagawa, Shun’ichi Kaneko, Hidenori Takauji · 2013
This paper presents a robust denoising model for high-frequency artifacts resulted by compressing images into JPEG. In this model, the authors used only simple evaluation value named Extended Three-valued Increment Sign (ETIS). ETIS represents the relationship of adjacent pixels, which one is brighter or almost the same. The authors expected that ETIS difference between Compressed Image and Noise Image would be small except edge region. Then they figured out the sum of the squares of those differences and utilized it in noise estimation. Only quantization process cause the artifacts, then they optimized DCT coefficient matrix in non-linearly based on ETIS, and estimated high-frequency artifacts as an independent approach without smoothing process. In the result, the model succeeded to reject noise with preservation of edge information. In addition, they compared the results with others those applied the traditional method called ε-filter and made sure that their method had similar or better improvement.