基于人工智能的图像自动渲染方法研究  

Design of automatic image rendering method based on artificial intelligence

作　　者：管阳[1] GUAN Yang(Shaanxi National Defense Industrial Vocational and Technical College,Xi’an 710300,China)

机构地区：[1]陕西国防工业职业技术学院,陕西西安710300

出　　处：《电子设计工程》2021年第3期157-161,共5页Electronic Design Engineering

基　　金：陕西省教育厅专项科研计划项目(19JK0081)。

摘　　要：针对传统图像渲染方法费时、复杂,无法满足现代设计应用场景的问题,利用人工智能技术中的卷积神经网络进行了图像自动渲染的研究。图像自动渲染算法主要分为两个部分:素材图像的特征提取与虚拟视图的渲染。通过卷积神经网络识别场景中图像的位移、缩放与其他形式扭曲不变性的二维图形轮廓,进而构建特征表征网络,用以提取图像特征并使之融合。同时,采用基于计算逻辑的图像校正算法来对齐不同视角所生成的虚拟视图,最终设计两层编码器-三层解码器的虚拟图像自动渲染模型。测试结果表明,所提出的图像自动渲染方法具有相对稳定的峰值噪声比与结构相识性,可以有效识别图像的特征并进行虚拟视图的渲染。In order to solve the problem that the traditional image rendering method is time-consuming and complex,which can not meet the needs of modern design of application scenes,the convolution neural network of artificial intelligence technology is used in the research of image automatic rendering.The automatic image rendering algorithm is mainly divided into two parts:the feature extraction of material image and the rendering of virtual view.The convolution neural network is used to recognize the displacement,scale and other distortion invariant two-dimensional contour of the image in the scene,and then the feature representation network is constructed to extract and fuse the image features.At the same time,the image correction algorithm based on computational logic is used to align the virtual views generated by different perspectives.Finally,a virtual image automatic rendering model of two-layer encoder and three-layer decoder is designed.The test results show that the proposed method has relatively stable peak to noise ratio and structure recognition,which can effectively identify the characteristics of the image and render the virtual image.

关 键 词：图像渲染 人工智能 卷积神经网络 计算逻辑 图像校正算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]