Human eye image enhancement method

Abstract

The embodiment of the invention discloses a human eye image enhancement method. A specific embodiment of the method comprises the following steps: acquiring a reference human eye image set and a target human eye image; performing human eye region segmentation on the reference human eye image set; performing pixel value sampling on the key area group set; constructing histogram vectors for the key region group set to obtain a histogram vector group set; performing data fitting on the histogram vector set to obtain a Gaussian mixture model set; carrying out resampling on the histogram vector group set to obtain a reference histogram vector set; performing human eye region segmentation on the target human eye image to obtain a target key region group; constructing a pixel histogram vector for the target key region group to obtain a pixel histogram vector group; mapping the pixel histogram vector set to a reference histogram vector set; and enhancing the pixel value of the target key area group to obtain an enhanced human eye image. According to the embodiment, quick and real-time human eye image enhancement can be realized.

Description

technical field [0001] Embodiments of the present disclosure relate to the technical field of data enhancement, and in particular to a method for enhancing human eye images. Background technique [0002] As a novel way of human-computer interaction, gaze tracking technology is of great significance for understanding user behavior intentions. The current gaze tracking technology is mainly divided into traditional methods and methods based on deep learning. The method based on deep learning needs to learn a large number of training samples to segment the pupil, iris, and sclera regions of the human eye, but the real human eye dataset is expensive to label. Although some human eye datasets have richer labels, other The collection scene is relatively single, and the performance of the neural network is not good when testing across data sets. In addition, due to the gap between the distribution of the virtual image and the real image, such as the blurred boundary of the iris an...

AI Technical Summary

Problems solved by technology

The method based on deep learning needs to learn a large number of training samples to segment the pupil, iris, and sclera regions of the human eye, but the real human eye dataset is expensive to label. Although some human eye datasets have richer labels, other The collection scene is relatively single, and the performance of the neural network is not good when testing across data sets

In addition, due to the gap between the distribution of the virtual image and the real image, such as the blurred boundary of the iris and sclera of the real human eye image, the virtual image is not realistic enough, etc., the neural network only learns the details that exist in the virtual image, but cannot be used on the real image. good generalization

Images