Proceedings Article | 20 December 2019
KEYWORDS: Resolution enhancement technologies, Image resolution, Image processing, RGB color model, Associative arrays, Reconstruction algorithms, Colorimetry, Color difference, Lawrencium, Image enhancement
Super resolution (SR) reconstruction and pixel interpolation are profitable technologies to acquire high resolution (HR) images from low resolution images. However, implementing the same interpolation or SR algorithms in different color spaces may still produce diverse results. Therefore, this study aimed to systematically investigate how the selection of color spaces take effect in the process of increasing image resolution. The resolution enhancement means involved an SR algorithm based on classified dictionary learning proposed by us and an SR algorithm based on deep learning with convolutional neural networks, as well as three typical pixel interpolation algorithms of bicubic, bilinear, and nearest. The evaluated color spaces involved RGB, YCbCr, YIQ, HSV, HSI, and CIELAB, which produced corresponding color coordinate systems. Based on the numerical measures of the peak signal to noise ratio (PSNR) and the color difference formula CIEDE2000 calculated between the original HR images and the processed versions, the results indicate that, YCbCr, YIQ, and CIELAB are suitable mapping spaces for resolution enhancement operations, and only the coordinate of bright and dark information is the dimension that need to be reconstructed by SR methods. Besides, color spaces with perceptual parameters of hue, brightness/lightness, and colorfulness/chroman/saturation are not suitable neither for SR reconstruction nor for pixel interpolation, which would cause severe color distortions. Thus, for preferable image effect, the recommended strategies are implementing SR algorithms for merely L * coordinate of CIELAB space or merely Y coordinate of YCbCr and YIQ systems, while the other two coordinates use the bicubic interpolation algorithm.
