DCT-Based Color Image Denoising: Efficiency Analysis and Prediction

In practice, acquired color images are inevitably noisy, and filtering/denoising procedure is used to suppress the noise. Although numerous denoising techniques have been proposed, they are not universally efficient in all considered practical situations. There are also contradictory requirements to color image denoising and their priority can be different and strongly dependent on the situation at hand. This also complicates the choice of a proper filter. Color images can be filtered in a component-wise (e.g., R, G, and B components separately) and in 3D (vector) manner. The latter group of approaches usually produces better results but has certain shortcomings and is less developed. One more aspect is that filtering efficiency is often analyzed and compared using only standard metrics (criteria) often ignoring recently designed visual quality metrics. Finally, before starting applying image denoising, it is good to understand how efficient can it be and is it worth to perform such afiltering. Then, the task of predicting denoising efficiency becomes very interesting.