Compressed Measurement of Structural Similarity Index

Abstract

Image quality index corresponds to amount of degradation in an image. Structural Similarity Index(SSIM) is one such image quality index considered under this project. The statistical parameters required for the computation of SSIM are defined in spatial domain. The computations involved can be reduced when performed in compressed or reduced domain. In addition, as many image processing applications are based on compressed data, reconstructing inferior quality image for quality measurement can also be avoided. Compressed data can be achieved either at processing level (image compression algorithms) or at acquisition level (compressive sampling). A new algorithm called CM-SSIM was proposed to compute the SSIM from the compressed data. This is done through reconstructing data from reduced dimension to appropriate basis system and defining statistical parameters associated with the basis system. The proposed algorithm is validated by performing correlation analysis between the DMOS with acutal SSIM and CM-SSIM. The correlation factors considered for analysis are pearson, spearman and kendall tau. The results confirms that the proposed algorithm exist a good relationship with the DMOS actual SSIM and would be suitable for future real-time systems