Image resizing with maximum entropy algorithm

Subsampling technologies are frequently applied to resize digital images into a lower resolution for the purpose of preservation and transmission for photography, scientific analysis, and Internet applications. Image subsampling methods should be analyzed for color distortion and retention of usable details from the original image. However, the interpolation-based resizing methods change the color information and attenuate a specific range of high-frequency components of which the human visual system makes significant use.
The proposed maximum entropy algorithm provides that, after an image goes through a lossy channel, here called subsampling, informative pixels are retained by analyzing the neighboring pixels. The selected pixels are represented directly in the output image, and color information is therefore preserved. From subjective observation and object evaluation using the entropy, contrast, and PSNR, the maximum entropy algorithm effectively maintains important features and color information and demonstrates better performance than the interpolation-based methods in some applications. Furthermore, the computational expense is suitable for real-time implementation due to the geometrically limited areas and simple arithmetic.