Method for synthesizing face backlight picture

Abstract

The invention provides a method for synthesizing a face backlight picture. The method comprises the following steps: S1, preprocessing an image; s2, face feature point detection: S2.1, a normal face image is sent to a face detection model, and face contour points are detected; s2.2, inputting the background image into a face detection model, and detecting background face contour points; s3, generating a closed face contour line; s4, mask construction: S4.1, applying a flooding filling algorithm to the image template M to generate a face region mask 0; s4.2, according to the corresponding contour points, solving homography matrixes of the two images; homography transformation is carried out on the mask0; a mask 1 after transformation is obtained; homography transformation is carried out on the face image; a transformed face image is obtained; s4.3, carrying out negation operation on the mask 1, and generating a background region mask 2; s5, target area extraction: S5.1, performing mask operation on the transformed face image and a mask 1, and extracting a face area to obtain an image I1; s5.2, performing mask operation on the background image and a mask 2, and extracting a backlight background area to obtain an image I2; and S6, image splicing: carrying out addition operation on the obtained image pixels to obtain a spliced image.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for synthesizing backlit pictures of human faces. Background technique [0002] The existing face recognition technology is essentially to establish a face recognition model, extract the feature vector of the face based on the model, and match the new picture ("new person") with the feature vector in the corpus, so that Determine the identity information of the "new person". The face recognition model based on deep learning requires a large number of face data sets to continuously train the model, so that the feature vector extracted by the model can better represent the face features. Moreover, the data set for training the face recognition model not only requires a large volume, but also needs to provide accurate labels and be compatible with various application scenarios (that is, face pictures in unrestricted scenarios). [0003] A restricted scene refers ...

AI Technical Summary

Problems solved by technology

[0004] At present, the large-scale face training sets publicly available on the Internet basically do not contain or include very few faces collected in backlit environments, and conventional data enhancement cannot produce the effect of face images collected in backlit environments. data set, will consume a lot of manpower and financial resources

Images