Evaluating Noise-Induced Quality Degradation During Wireless Channel Transmission of Laparoscopic Videos

Noise-induced degradation of wirelessly transmitted laparoscopic videos significantly compromise surgical precision by reducing perceptual clarity and impairing depth perception. To address this, objective video quality assessment (VQA) methods can be used to develop automated quality control systems capable of mitigating the impact of noise. This paper presents a no-reference laparoscopic VQA method that captures the interdependencies between depth and perceptual features to mimic human perception of noisy laparoscopic videos. Framewise extraction of luminance, local binary pattern, motion and depth maps are performed, with depth-based pairs used to compute correlation maps for representing feature interdependencies. These maps are spectrally decomposed and modelled using generalized Gaussian distribution parameters at the frame level. The model parameters are then used with subjective labels to train a long-short term memory framework, and the mean of the predicted frame qualities are computed as the final video quality. The results demonstrate that both experts and nonexperts rely heavily on interdependent luminance and depth features for evaluating video quality, while experts also show a strong dependence on texture-depth interactions. The proposed method outperforms other state-of-the-art objective methods, underscoring the role of depth perception in replicating real-life human vision.