Quality assessment of color images affected by transmission error, quantization noise, and noneccentricity pattern noise

Abstract

Security images transmitted over network or wireless mediums are subjected to a variety of transmission errors and non-eccentricity pattern noise. Some of these errors are due to hardware limitations that can result in lower quality images caused by quantization noise. Currently, there are no quantitative metrics to evaluate the visual quality of images containing transmission errors based on human perception. Traditional methods rely on human evaluators, which is not always feasible in live field security applications. Image quality measures are desired to evaluate these images and to help with parameter selection in their enhancement process. The performances of existing spatial domain image quality measures are sensitive to the image contents and attributes. The image quality measures that operate in the transform domain such as the DCT, DFT, or DWT, are independent of image attributes such as periodicity, texture, and randomness; however, these measures are mostly contrast measures and designed to handle gray scale images. Currently, there are no transform domain measures for color images that are suitable for these classes of noise and distortions. In this article, we introduce a no-parameter, no-reference transform domain measure of enhancement for color images. The test images are distorted with JPEG/JPEG2000 transmission errors, quantization noise, and non-eccentricity pattern noise. Our measure utilizes the spectrum of the color content of images in the transform domain. This measure can help optimize parameter selection during the enhancement process or to autonomously evaluate images in accordance with human visual perception.