Image aesthetic enhancement method based on deep neural network and cascade regression

Abstract

Embodiments of the present invention provide a method for enhancing image aesthetics based on deep neural network and cascaded regression, comprising the following steps: S1, providing an image with a size of C 0 The original image is solved by the cascade regression algorithm to obtain the minimization target; S2, the image feature extraction function of the deep convolutional neural network is trained through the AVA data set and the CHUKPQ data set; S3, the deep feature x is extracted through the deep convolutional neural network t , and through the spatial pyramid pooling layer of the deep convolutional neural network, the size is C 0 The original image is transformed into a feature vector of (2*2+3*3+4*4)*32; S4, the extracted depth feature x t It is applied to the random fern regressor, and the gradient descent method is used to learn the elementary regressor, and the candidate result C is output through the cascaded regressor j (1≤j≤4); keep iterating until the error no longer decreases, and obtain the minimum objective function of the primitive regressor; S5, combine the primitive regressor obtained in step S4 with the depth feature x t Perform T cycles in an iterative manner, and obtain the final cropped image through multi-stage cyclic cutting.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to an image aesthetic enhancement method based on deep neural network and cascaded regression. Background technique [0002] An important source of human perception of the world is through image information. Studies have shown that about 80% to 90% of the information that humans obtain from the outside world comes from image information obtained by human eyes. The understanding of image information includes not only common computer vision tasks such as image classification, object detection, and object tracking, but also the understanding of semantic and aesthetic information of images. Understanding the aesthetic information of images has rich applications, such as providing pleasing aesthetic images to users in image repositories. General consumers or designers can take advantage of automated image beauty enhancement systems to make better decisions. Therefore, it is of ...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the purpose of the embodiments of the present invention is to provide an image aesthetic enhancement method based on deep neural network and cascaded regression, so as to improve the poor aesthetic enhancement effect caused by the lack of data marks in the prior art and the image being too focused on the original image. question

Images