A semi-automatic face key point labeling method and storage medium

Abstract

The invention relates to a semi-automatic human face key point labeling method, the method comprising the following steps: dividing a human face picture to be marked into two parts for labeling respectively, a human face feature part and a human face contour part; extracting human face facial features Part of the texture edge is used as the target curve V; anchor points P are provided for points that deviate from the texture edge of the facial features A , register the key points on the target curve V until the satisfactory labeling result of the facial features is obtained; extract the texture edge of the contour part of the face as the target curve V′; register the key points on the target curve V′ until The satisfactory annotation results of the face contour part are obtained. Different from the prior art, the present invention can automatically calculate the position of most of the key points by the labeling tool under the condition that the labeler provides a small amount of key information. Different labeling methods are designed for different parts, and it has strong robustness for the initialization of key point labeling and has a wide range of applications.

Description

technical field [0001] The present invention relates to the technical field of image processing and pattern recognition, in particular to a semi-automatic face key point data labeling method and storage medium, especially a semi-automatic face recognition method based on iterative non-rigid closest point registration method and three-dimensional variable model Dense Keypoint Labeling Method. Background technique [0002] The key points of a dense face are a series of points on the face that have fixed semantics or describe specific contours, such as the chin, the tip of the eyebrows, the corners of the eyes, and around the eyes. Dense face keypoint location is an important preprocessing step in many computer vision tasks based on face understanding, such as face recognition, 3D face reconstruction, and face pose estimation. Existing face key point positioning algorithms, especially key point positioning algorithms based on deep learning, have a strong dependence on data, so...

AI Technical Summary

Problems solved by technology

This method is more feasible when there are fewer key points, but it is time-consuming when applied to dense key points

The existing challenges mainly include the following two aspects: First, in traditional labeling methods, for points without clear semantic positions (such as contour points), the final position depends too much on the subjective judgment of the labeler, and labeling often requires multiple Completed by the annotator, which leads to random annotation positions and poor annotation quality of the dataset

Secondly, repeated drag and drop confirmation requires the annotator to continue to concentrate at a high level, which will cause the annotator to quickly enter fatigue and lead to a rapid decline in annotation efficiency

Images