1. Demand Analysis
There is a word called "Skin Ruo Ningzhi", literally translated as skin like solidified oil, describing the skin as white and smooth, which is a general evaluation of beautiful women. In real life, our skin has some pores, spots, etc., which are opposite to the above-mentioned "smoothness" in appearance, so skin resurfacing has also become a basic and important function in the beauty algorithm. Making the skin smoother is the goal of the dermabrasion function.
2. Quantitative analysis
Our goal is to make the skin smoother, so 1) we need to eliminate the influence of "noise" such as spots and pores, and 2) we also need to preserve the sharpness of areas such as facial features; this actually belongs to the field of image processing. The function to be realized by the noise algorithm. This type of algorithm includes bilateral filtering, guided filtering, etc. In the following, we will introduce around bilateral filtering, and then introduce another edge-preserving and noise-reduction algorithm with a smaller amount of calculation.
3. Image filtering
3.1 Mean filtering
The filtering when the value of each element in the convolution kernel is the same is mean filtering, as shown in the figure below, it has the following properties: 1) The resulting image is blurred; 2) The noise and the edge are both affected by the same treatment; 3) The larger the convolution size Larger, more blurred.
