Flattening of New Media Design Based on Deep Reinforcement Learning

Abstract

In order to strengthen the flat design of new media, this paper proposes the research on the flat design of new media based on deep reinforcement learning. Firstly, this paper introduces the basic principle of deep belief network and lists the flat design methods of new media animation. An algorithm based on deep reinforcement learning is designed, which can gradually fill the missing area image. A set of repair strategy is designed through the new media animation image repair algorithm. Taking animation facial expression recognition Based on deep learning as an example, this paper expounds the theme through the combination of theory and practice. In the Jaffe database and Cohn Kanade database, three new media animations with different resolutions of $16\times 16$ , $32\times 32$ , and $64\times 64$ are taken as examples, and DBN (Deep Confidence Network) method is compared with other five common classification methods. DBN has the highest identified value. It should be noted that the accuracy of the DBN method is $23.79\%$ , $4.31\%$ , and $4.80\%$ higher than that of MLP under $16\times 16$ , $32\times 32$ , and $64\times 64$ , respectively. The recognition performance of DBNs is almost the highest. Although the classical SVM method has achieved 98.11% recognition rate on $32\times 32$ image, which is higher than the DBN method, the fluctuation of the SVM method is relatively large, and the recognition rate on $16\times 16$ image has declined greatly. Relatively speaking, the recognition rate of the DBN method is relatively stable.