Pupil edge identification method and system based on unet pseudo twin network

Abstract

The invention provides a pupil edge identification method and system based on a unet pseudo twin network. The method comprises the following steps: inputting eye image data into a cross-model feature fusion network; wherein the cross-model feature fusion network comprises two sub-networks, each sub-network comprises an encoder and a decoder, an output result of the encoder of each sub-network in the two sub-networks is fused with an output result of the decoder of the other sub-network, the encoder extracts feature information of different scales from an image, and the decoder extracts feature information of different scales from the image, the decoder carries out cross-network fusion output on the feature information; and processing a feature map output by the cross-model feature fusion network, carrying out edge detection, and fusing a result with an original image. According to the method, the features are subjected to cross-network fusion, and the features from different coding stages are aggregated, so that the network can maintain detail information of an original image, the expression ability of the network is enhanced, and the pupil edge recognition precision is improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a pupil edge recognition method and system based on a unet-based pseudo twin network. Background technique [0002] Research related to the human eye involves multiple disciplines such as artificial intelligence, cognitive psychology, virtual reality, image processing, etc. The application of the human eye involves identification, medical diagnosis, virtual technology, sentiment analysis, etc. Many of these applications depend on the iris and its central pupil. Under ideal conditions, professional equipment is often used in order to collect eye information such as fixation time, pupil diameter and other data. In reality, although the boundary between the pupil and the iris is very clear, due to the influence of light distribution, occlusion of eyelids and eyelashes, and light spots, the eye image acquired by ordinary equipment is not as clear as it is under ideal conditions. At ...

AI Technical Summary

Problems solved by technology

The deformable template method can also identify pupils, but this method is computationally complex, has a large amount of calculation, and slow processing speed. The selection of initial parameters and positions of the template is also a restrictive factor for this method.

The iris or pupil can be identified by directly using the Canny edge detection algorithm, but the threshold of edge segmentation is related to the viewing angle direction and face posture. If a fixed threshold is used for processing, it is difficult to obtain the desired effect

Images