Enhancing Laparoscopic Video Quality Assessment: A Model Addressing Sensor and Channel Distortions

This letter presents a supervised quality assessment (QA) model for medical laparoscopic videos (LV) affected by sensor and channel distortions. The quality discerning features of LVs are computed based on the statistical properties of the singular, left, and right eigenmatrices of the singular value decomposition (SVD). Specifically, inspired by the perceptual characteristics of the human visual system, the LV frames are decomposed into multiple subbands using a two-level discrete Haar wavelet transform. Then, the maximum value of SVD matrices across columns of each decomposed subband is computed to estimate the frame level quality discriminative features of the LV. Further, the frame level features are averaged over the number of frames to estimate the quality aware feature set of the LV. The top-ranked features computed from the ReliefF algorithm and the expert opinion subjective scores are given to the support vector regressor to estimate the quality score of a test LV. Experimental results on the LVQA dataset demonstrated that predictions of the proposed LVQA model correlate well with expert subjective ratings and outperformed the performance numbers of existing image and video QA models.