Marking method of background of marked image and skin problem identification method and device

Abstract

The invention relates to the technical field of image recognition, in particular to a marking method of a background of a marked image, a training method of an artificial neural network for skin problem recognition, a skin problem recognition method / device and a computer readable storage medium. According to the marking method of the background of the marked image, the skin image marked with the marked area and the background area is further processed, the area adjacent to the marked area in the background area is marked as the adjacent area, and the weights of the adjacent area and the remaining non-adjacent areas in the background area are marked respectively; the skin images marked through the marking method serve as a training set and a verification set to train a multi-task convolutional neural network model, and finally the skin images to be recognized are input into the trained skin problem recognition network to be recognized, so that the marked areas and the skin problem types in the marked information of the marked areas are obtained. And skin problem identification with low background misidentification probability is realized.

Description

technical field [0001] The present invention relates to the technical field of image recognition, in particular to a marking method for the background of marked images, a training method for artificial neural networks for skin problem identification realized based on the marking method, and skin problems trained by the training method A skin problem identification method / device implemented by a recognition network, and a computer-readable storage medium. When the computer-readable storage medium is executed by a processor, the above-mentioned marking method, and / or training method, and / or skin problem identification method can be realized. Background technique [0002] Existing technologies utilize pre-trained artificial neural networks to automatically identify skin problems such as freckles, age spots, acne, and moles on human face images. To this end, it is necessary to manually label a large number of images with skin problems, that is, to mark out the areas with skin pr...

AI Technical Summary

Problems solved by technology

However, on real human faces, multiple skin problems often exist at the same time, such as age spots A, acne B, mole C, etc. When identifying skin problems on a face image, if only the marked age spots A are identified, then, During training, the artificial neural network will regard acne B and mole C as background areas and mistake them for normal skin, and then continue to train the artificial neural network to recognize acne B / mole C, which will be affected by the previous misidentification

The current solution is to reduce the weight of the background during training. For example, the weight of the currently recognized skin problem age spot A accounts for 99%, while the background area containing sunburn B, pigmentation C and other skin problems only accounts for 1%. This can Reduce the interference caused by other skin problems such as sunburn B and pigmentation C to the background recognition, but it will also lead to insufficient recognition weight of the background area, reduce the background recognition ability and lead to background misrecognition

Images