Image identification method and system

Abstract

The application discloses an image identification method and system. The image identification method comprises the steps of obtaining the information of the images, and extracting the characteristic data of the images; marking the characteristic data, and converting into the expression form of multiple examples; carrying out the machine learning via a multi-example weighting packet, and extending a multi-task learning environment; training to obtain a multi-example multi-task image identification classifier, and identifying and classifying the images. According to the above image identification method disclosed by the application, and by utilizing the multi-example weighting packet, the influence of the noise in the image identification on the classification results is reduced, at the same time, the classification precision is improved. Moreover, an algorithm extends to the multi-task environment, and by utilizing the advantages of the multi-task environment, the correlation between the images can be utilized effectively, and the identification accuracy is improved further.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to an image recognition method and system. Background technique [0002] At present, with the rapid development of the Internet, the network data is getting bigger and bigger, especially the image data is increasing geometrically. Therefore, how to use these data reasonably and effectively has become our thinking. At the same time, with the great development of machine learning, online image recognition, such as handwritten numbers, face recognition, etc., has also been widely used. [0003] On the one hand, in the process of application, due to the limitation of handwriting equipment, the error of shooting equipment, the influence of uncertain factors such as data environment and transmission process, the obtained images often contain noise data, and the validity of the data is lacking. Support vector machines based on multiple examples cannot effectively learn and expr...

AI Technical Summary

Problems solved by technology

[0003] On the one hand, in the process of application, due to the limitation of handwriting equipment, the error of shooting equipment, the influence of uncertain factors such as data environment and transmission process, the obtained images often contain noise data, and the validity of the data is lacking. Support vector machines based on multiple examples cannot effectively learn and express, thereby reducing the recognition accuracy of images and seriously affecting the wide application of image recognition; on the other hand, due to the insufficient image data acquired in image recognition applications, And when these images are relatively similar and have potential connections, the support vector machine based on single-task multi-example cannot further utilize the connection between images, thus affecting the efficiency of recognition, reducing the accuracy of image recognition and making image recognition more accurate. rate is not high

Images