Color Transformation for Compressive Computing in Image Filtering

In this paper, we propose an approximated acceleration framework for image filtering by subsampling with color space transformation. Image subsampling then processing is a simple acceleration approach for general image processing; however, the processing loses the resulting accuracy. To keep the high accuracy, the proposed framework converts a color space of the input image, i.e., as RGB, into an appropriate color space, such as YUV, DCT, and PCA. These kinds of the color spaces gather important signals to one channel. The channel information is biased; thus, we downsample other non-important signals, and then we perform filtering for the converted full-sample and downsample signals. The experimental results show that the proposed framework achieves high accuracy and fast computational time for various image filters.