Abstract
This paper proposes an objective distortion measure for binary images. Whereas most previously proposed measures decide the distortion weight of a distorted pixel based on the relationship with the neighbouring pixels in the spatial domain, we observe the characteristics of distorted pixels in multi-spectral spatial-frequency bands. Similarities between the binary wavelet transform and the human visual system are utilised to obtain the distortion weight. When we decide the distortion weight of a distorted pixel, we consider flatness, edge orientation and texture complexity of the distorted pixel at each level of binary wavelet decomposition. Experimental results on various sample images show that the proposed measure is more adequate to human than previously proposed measures.
This work was supported by the Agency for Defense Development of Korea.