Method used to identify lipstick number

Abstract

The invention discloses a method for identifying lipstick numbers, comprising the following steps: obtaining a training picture; preprocessing the training picture; training a decision forest through the preprocessed training picture, and the decision forest includes a plurality of decision trees; The preprocessed training picture trains the SVM classifier; obtains the picture to be recognized; preprocesses the picture to be recognized; inputs the preprocessed picture to be recognized into the trained SVM classifier and decision forest to obtain the judgment result; the SVM classifier Vote with the judgment result of the decision forest to get the final recognition result. The benefit of the present invention is that the provided method for identifying the lipstick number converts the picture into hsv, which removes the interference of the shape and brightness on the recognition result, and simultaneously uses the SVM classifier and the decision forest for comprehensive judgment, which improves the mouth color. Accuracy of red number recognition.

Description

technical field [0001] The invention relates to a method for identifying lipstick numbers. Background technique [0002] Most of the existing lipstick number classification methods use deep learning to train a convolutional neural network model to identify the lipstick number. The disadvantage of the existing method is that the features extracted by the convolution method are shape features, while the lipstick’s Classification relies on color features, and the results of classification based on shape are very inaccurate, and the number of neural network layers using deep learning is generally too deep, and it is extremely easy to overfit for data with minimal differences and insufficient quantities. It is very accurate, but the untrained data is very inaccurate, and if the number of network layers is not deep, the ability to distinguish pictures with small differences is extremely weak. Contents of the invention [0003] The invention provides a method for identifying lip...

AI Technical Summary

Problems solved by technology

[0002] Most of the existing lipstick number classification methods use deep learning to train a convolutional neural network model to identify the lipstick number. The disadvantage of the existing method is that the features extracted by the convolution method are shape features, while the lipstick’s Classification relies on color features, and the results of classification based on shape are very inaccurate, and the number of neural network layers using deep learning is generally too deep, and it is extremely easy to overfit for data with minimal differences and insufficient quantities. Very accurate, but untrained data is very inaccurate, and if the number of network layers is not deep, the ability to distinguish pictures with small differences is extremely weak

Images