Enhancing Surgical Laparoscopic Video Quality Assessment With Integrated Feature Fusion Accounting for Sensor and Transmission Distortions

Abstract

In this letter, an opinion-aware quality assessment (QA) model for surgical laparoscopic videos (LVs) considering sensor and transmission distortions is proposed based on statistical disparities between luminance and color components of the opponent color space (OCS). First, the luminance variations among the frames of distorted LVs are computed based on the energy of the Gabor subbands and weighted histogram features of the local binary pattern map. Second, the color degradations of each frame of LV are estimated based on the chromatic components of the OCS using moment statistics and the shape and spread parameters of the asymmetric generalized Gaussian distribution. These features are computed across two scales, concatenated, and pooled to obtain the overall quality representative feature set of the LVs. Finally, an AdaBoost back propagation neural network is utilized to map the extracted feature set to quality scores using labels as surgeons opinion scores. Extensive experiments demarcate that the proposed QA model for surgical LVs outperforms the existing video QA models with an overall linear correlation coefficient of 0.9800 and Spearman rank order correlation of 0.9247 on the LVQA dataset, respectively.