Marc Olano, CSEE Department of Computer Science and Electrical Engineering
Wesley Griffin, CSEE Department of Computer Science and Electrical Engineering
While grayscale image quality assessment is well-understood, color image quality assessment remains an open research problem. For many applications, using a grayscale image quality metric is sufficient and there are excellent existing algorithms. There are cases, however, when color information must be considered for image quality. One such application is texture compression. Modern image compression methods leverage the sensitivity of the human visual subsystem to vary the compression of the luminance and chrominance components of an image. A modern compression algorithm could use an image quality metric to drive compression rates, but if the metric does not handle color, then the chrominance components would be heavily compressed, compromising quality.
There are several existing algorithms for color image quality assessment. These algorithms tend to be complex and use Contrast Sensitivity Functions to model the sensitivity of the Human Visual System (HVS). Recent advances in grayscale image quality assessment have resulted in very simple formulations for image quality. These “top-down” approaches attempt to simulate the HVS response instead of modeling the HVS. One such approach is the Structural Similarity Index Metric (SSIM). Extending SSIM to work with color would provide applications with a simple but accurate color image quality metric. We present a new algorithm that combines SSIM with CIELAB color differencing to perform objective image quality assessment on color images.