Human eye detection and localization method based on deep autoencoder

Abstract

The invention discloses a method for detecting and locating human eyes, comprising: for all images in a training set with calibrated rectangular frame positions of human eyes, using the rectangular frame positions of human eyes to generate a binarized label map; Randomly select small image blocks on the image, unsupervised layered training of multiple self-encoders to build a deep self-encoder, and use the weights of each layer in the self-encoder to initialize the deep self-encoder; in the original image and label map Randomly select small original image blocks and small label image blocks at the same position of , use the small label image blocks as supervision information, and use the small original image blocks as input to optimize the depth autoencoder; generate multiple For small image blocks to be tested, use the deep self-encoder to obtain the small image blocks to be tested for each small image block to be tested and merge them together to obtain the image to be tested. The position of the human eye can be obtained by using coordinate projection or finding the contour.

Description

Technical field [0001] The invention relates to the field of pattern recognition and machine learning, in particular to image target detection. More specifically, the present invention relates to a human eye detection and positioning method based on a deep autoencoder. Background technique [0002] The explosive growth of the application of biometric recognition technology and the huge demand for migration of biometric recognition algorithms to embedded and mobile platforms have made rapid eye detection and positioning increasingly important. The traditional target detection algorithm builds a feature pyramid of the image, and extracts windows slidingly on the pyramid, classifies the extracted windows, and finally obtains the position of the target through a non-maximum suppression operation. Although this method has achieved good detection results in a simple background, and can achieve real-time on an ordinary computer, this method cannot handle target detection in a complex b...

AI Technical Summary

Problems solved by technology

Although this method has achieved good detection results in simple backgrounds and can achieve real-time performance on ordinary computers, this method cannot handle target detection in complex backgrounds well, as well as objects with deformation, viewing angle changes and occlusions. problems, and difficult to achieve real-time on embedded and mobile platforms

In addition, the current target detection method based on the deep convolutional neural network has achieved high detection accuracy, and can handle complex backgrounds and problems with deformation and viewing angle changes very well. However, due to its huge amount of calculation, even with the help of parallel computing technology, it is also difficult to meet the real-time requirements

Images