Human binocular color fusion model based on BP Neural Networks prediction

Abstract

Stereoscopic display vision is significantly impacted by the color distortion of left and right eye images. When the human eye receives a specific range of dissimilar color information separately, the visual system combines them into a single color through binocular color fusion. In this study, we present experimental findings which compare the accuracy of a common binocular color-fusion model that was trained utilizing both linear fitting and back-propagation neural networks. Patient binocular color contrast test data was collected by eye care professionals working in private eye clinics. The results indicated that the back-propagation neural network produced RMSE errors of 0.9819 and 0.9662 for predicting binocular contrast, which were superior to the linear fitting method with errors of approximately 0.5. The BP neural network algorithm employed demonstrates predictive capabilities and lessens the occurrence of color redundancy. This reduction in redundancy holds the potential to decrease expenses associated with stereo imaging in future applications.