Automatic colorization of grayscale image using SVM
For the image colorization, the following steps are performed:
- Conversion of RGB image to LAB color space
- Use of K-means clustering to reduce the number of image colors e.g. to 64.
- Image segmentation based on superpixels using SLIC algorithm
- SURF and Gabor feature extraction
- Train and use of SVM for color prediction
- Optimization using GraphCut algorithm
In the following image the original image alongside the SVM and GraphCut output are compared using 49 (α) and 304 (β) superpixels, respectively:
Additional results are shown here: