shimizuwasedau / grayscale-oil-painting-filter Goto Github PK

Grayscale oil painting filter

1. Description of grayscale oil painting filter In this project, I developed a grayscale oil painting filter which can change a picture from RGB space into a grayscale picture in the form of oil painting.

2. Technology used Outline: In the project, I used color space conversion from BGR grayscale, medianBlur, modified Laplacian filter. Those functions are realized by myself instead of using opencv functions. 2.1 toGray function First, I use toGray() function to transform original image from RGB color space to grayscale space: Here I used I=(11B+59G+30*R)/100, as you can see in the expression, I used integer instead of float in order to reduce computation and speed up the program.

2.2 medianBlur function In the program, medianBlur() function is self-realized median filter which uses a 77 window to filter the matrix: Firstly, we use a 77 matrix to filter the picture pixels, then we use a function called medValue() function to get the median value of the pixels within this window, and then use the median value to replace each pixel in the window. This medValue() function is realized by the bubble sorting function to find the median value the 7*7 matrix. Here we use median filter because normal picture in the Internet contains a lot of noise, a median filter can reduce the effect of salt and pepper noise very well.

2.3 modified Laplacian function Laplacian() function is a modified Laplacian filter which also uses convolution but with an different kernel matrix: Usually, in Laplacian filter of opencv, we use 33, 55 or 77 kernel matrix, for example, if we want to use a 55 kernel, it is listed as bellow: 0 0 1 0 0 0 1 2 1 0 1 2 -17 2 1 0 1 2 1 0 0 0 1 0 0 Table 1 5*5 laplacian filter If we use the table 1 filter as the kernel, we will get the edge of the grayscale image of the original image, but here we use a different kernel which is shown bellow:

0 0 -1 0 0 0 -1 -2 -1 0 -1 -2 17 -2 -1 0 -1 -2 -1 0 0 0 -1 0 0 Table 2modified laplacian filter We will get a grayscale oil painting image instead of the edge image of the original picture.

3. Summary My project is a grayscale oil painting filter which is realized by modified Laplacian filter. Instead of using laplacian filter as edge detector, here we use it as a different filter which transform the picture into a different style by converting the value of the kernel matri.

4. Future work
   In this project, we can only get a grayscale oil painting picture, the function doesn¡¯t work well in producing an oil painting picture with RGB color space. Next, I will try find a corresponding filter which works as well with RGB pictures.

Reference

1. Laplacian filter: http://www.uoguelph.ca/~hydrogeo/Whitebox/Help/FilterLaplacian.html
2. Spatial Filters: http://homepages.inf.ed.ac.uk/rbf/HIPR2/log.htm