Image processing method and system, and server

Abstract

The embodiment of the invention discloses an image processing method and system, and a server. The method comprises the following steps of: obtaining a face image to be processed; inputting the face image into a preset convolutional neural network model provided with a loss function, and performing parameter regularization processing of the loss function to allow the convolutional neural network model to perform directed screening of between-class distances after image classification is enlarged; and obtaining classification data output by the convolutional neural network model, and performingcontent understanding of the face image according to the classification data. The parameter regularization processing is added on the basis of the loss function of the convolutional neural network model to allow each class extracted by the convolutional neural network model to be close to the weight of each class itself in directions, and therefore, it is ensured that the extracted face image features can stay the same as much as possible in a cosine space, and intra-class features of the face image features have better convergence.

Description

technical field [0001] Embodiments of the present invention relate to the field of image processing, in particular to an image processing method, system and server. Background technique [0002] Face recognition refers to the technology of using computers to process, analyze and understand face images to identify targets and objects in various face images. Face recognition can be applied in many fields such as security and finance. The process of face recognition is generally divided into three stages: face detection, face alignment, face feature extraction and comparison, and face feature extraction is the process of face recognition. key technologies. [0003] With the development of deep learning technology, the convolutional neural network has become a powerful tool for extracting face features. For the convolutional neural network with a fixed model, the core technology is how to design the loss function so that it can effectively supervise the convolutional neural net...

AI Technical Summary

Problems solved by technology

[0004] However, the inventors of the present invention found in the research that the difficulty of the Triplet Loss function method lies in the preparation of data. The loss function is essentially a sampling process. The experimental data given by Google is 260 million, which is generally difficult to meet

The feature extraction method of Softmax's cross-entropy loss function is a non-end-to-end method, which is simple and easy to implement, but due to the lack of restraint of the intra-class data obtained by its training, the inter-class distance that determines the size of the cosine space distance is not obvious enough. , the classification data is poorly discrete, resulting in the inability to improve the accuracy of content understanding

Images