基于混合特征提取的PatchMatch图像风格迁移算法研究与应用

Research and Application of PatchMatch Image Style Transfer Algorithm Based on Mixed Feature Extraction

图像风格迁移算法是当前计算机视觉领域的研究热点。传统图像风格迁移算法通常从风格图中采集纹理并将其迁移到内容图中,该类算法在对色彩与纹理较为复杂的图像进行处理时,合成效果不理想,且运算速度较慢。随着人工智能技术的不断发展,图像的艺术风格化已可使用卷积神经网络完成。但如何实现参数的合理调整,以及如何简化预训练模型,该类算法仍存在一定的局限性。为解决上述问题,文章提出基于颜色特征,基于LBP纹理特征,基于结构张量特征的混合特征提取新算法,并结合运用PatchMatch算法来解除这些限制,最终进一步通过GPU实现实时的图像风格迁移算法。

Image style transfer algorithm is a research hotspot in the field of computer vision.Traditional image style transfer algorithms usually collect textures from the style map and migrate them to the content map.When these algorithms process images with complex colors and textures,the synthesis efect is not ideal and the operation speed is slow.With the continuous development of artificial intelligence technology,the artistic stylization of images can be completed using convolutional neural networks.However,such algorithms still have some limitations on how to achieve reasonable adjustment of parameters and how to simplify the pre-training model.In order to solve the above problems,this paper proposes a new hybrid feature extraction algorithm based on color features,LBP texture features,and structure tensor features,and uses PatchMatch algorithm to remove these restrictions.Finally,the real-time image style migration algorithm is implemented through GPU.