基于小波预测的超分辨率素描人脸合成方法

Super-resolution sketch face synthesis via wavelet prediction

在使用人工智能算法等途径生成高质量的素描人脸图像时,若输入光学图像分辨率较低,生成的素描图像质量也会较差。在仅有低分辨率光学图像的条件下,提出一种基于小波预测的超分辨率素描人脸合成方法合成高质量的素描图像。在素描人脸合成网络的基础上引入超分辨率模块,通过对高分辨率图像的小波包分解系数进行预测,在端到端的框架下同时完成对图像的素描人脸图像合成以及超分辨率重建,提高合成图像的质量及分辨率。通过在CUHK学生数据集上与目前领先的超分辨率重建方法进行实验对比,该方法相较其它对比方法取得了更加优越的实验结果。

When using artificial intelligence algorithm and other approaches to generate high-quality sketch face images,if the optical image resolution is low,the generated sketch image’s quality will also decrease.A super-resolution sketch face synthesis method based on wavelet prediction was proposed to improve the quality of sketches generated from low-resolution optical images.The super-resolution module was introduced for the sketch face synthesis network.By predicting the wavelet packet decomposition coefficients of the high-resolution image,the sketch face image synthesis and the super-resolution reconstruction of the image were completed simultaneously under the end-to-end framework,thereby improving the quality and resolution of the synthesized image.Experiments were implemented on the CUHK student data set and were compared with the current leading super-resolution reconstruction methods.Experimental results are better than other methods.