Morphological image segmentation applied to video quality assessment

Abstract

This paper presents a methodology to morphological video segmentation used as part of a method to estimate the subjective video quality assessment. The segmentation goal is to partition the sequence frames in three different regions: homogeneous, border and texture. Previous work has shown the importance of computing video impairment measurements to each of these regions separately, in contrast to use global measurements. The segmentation approach presented in this work uses a collection of morphological tools such as connected smoothing filters, morphological gradients, and watershed. Special attention is devoted to describe the recent concept of connected filters, used as the kernel of the morphological segmentation algorithms. The performance of two morphological segmentation paradigms, one based on the flat zone, and the other based on the watershed-plus-markers approach are evaluated and compared to other segmentation methodology used in video quality assessment.