Deep learning instance segmentation-based stain acne detection and health evaluation method

Abstract

The invention discloses a deep learning instance segmentation-based stain acne detection and health evaluation method, and the method comprises the steps: collecting a front image of a stain acne human face, marking stains and acnes in the front human face, and forming a blocky human face health problem data set; training a stain acne instance segmentation model, training a backbone network from an Image Net image classification data set DCls by using a two-stage channel selection pre-training method, and training the instance segmentation model from a focus boundary segmentation data set DLBSpublished by ISIC; on the massive face health problem data set, selecting three optimal channels of the instance segmentation model, and training to obtain a pigmented acne instance segmentation model; collecting a front image of the face of the user, detecting and segmenting by the color spot acne instance segmentation model to obtain a color spot distribution diagram and an acne distribution diagram, and obtaining position and area data of color spots and acnes of the face of the user; and evaluating the health degree of the human face according to the number of the skin problems, the located human face subarea and the concentration degree.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an image processing method for identifying and evaluating acne pigmentation on human faces. Background technique [0002] In the medical cosmetology industry, pigmentation and acne are one of the most important problems in the treatment of facial skin problems. At present, domestic facial pigmentation and acne detection and identification methods mainly use artificial subjective judgment, lack of objective quantitative analysis, how to quickly and accurately Identifying the problem of facial skin pigmentation and acne has become an urgent problem to be solved. The detection methods about acne and pigmentation mentioned in existing patents mainly involve traditional image processing methods. For example, CN106529429A provides a pigmentation detection module, but Based on the watershed algorithm, the accuracy is not high, the environmental requirements are high, and the applicabilit...

AI Technical Summary

Problems solved by technology

This method mentions a deep learning-based acne identification method, but does not describe its specific detection method and counting scoring mode, and the method is not reproducible

Images