Complete extraction method of true color eye image graph white eye area

Abstract

The invention relates to a complete extraction method of a true color eye image graph white eye area. The method comprises the following steps of using a trained edge extraction model to carry out edge extraction of white-eye and black-eye complete areas on different-visual-angle 24-bit true color eye image graphs of natural light; after expansion and corrosion morphology conversion, acquiring a completely-enclosed edge binary image, calculating a maximum connection area, mapping the binary image to an original 24-bit true color eye image graph and acquiring the white-eye and black-eye complete areas; using a trained black-eye position prediction model to calculate position coordinate information of a black-eye external connection rectangle, and rapidly positioning a black-eye position; in the white-eye and black-eye complete areas, using the black-eye position coordinate information to roughly segment a black-eye area, using a Hough detection circle to calculate and acquire a black-eye center and a radius, and accurately segment the black-eye area; and according to visual angle orientation information of an eye image and the acquired black-eye area, removing a black-eye area image in the white-eye and black-eye complete areas and acquiring a white-eye area. In the invention, the white-eye area can be accurately, effectively and completely extracted.

Description

technical field [0001] The invention relates to a method for extracting an eye image map, in particular to a method for completely extracting the white eye region of a true-color eye image map. Background technique [0002] In the past year, deep convolutional neural networks have had epoch-making applications in image and speech recognition technology, especially in the direction of medical image prediction and diagnosis. Various achievements have been published. Medical image images are mainly grayscale images. Sclera recognition, iris recognition and fundus image recognition are mainly aimed at grayscale images and false color images. For the eye images that deal with natural light, we mainly extract and analyze the features in the eye images of the elderly and those with serious illnesses. However, when using various edge detection algorithms, the problem of edge disconnection or inappropriate threshold will occur, and the edge of the white eye extracted by traditional ...

AI Technical Summary

Problems solved by technology

However, when using various edge detection algorithms, the problem of edge disconnection or inappropriate threshold will occur, and the edge of the white eye extracted by traditional convolution training is affected by blur or occlusion and is incomplete.

Images