基于改进神经网络的4K超高清图像清晰化技术研究  

Research on 4K ultrahigh-definition image clearness technology based on improved neural network

作　　者：黄卫[1] 文坤梅[2] HUANG Wei;WEN Kunmei(Neijiang Vocational and Technical College,Neijiang 641100,China;Huazhong University of Science and Technology,Wuhan 430000,China)

机构地区：[1]内江职业技术学院,四川内江641100 [2]华中科技大学,湖北武汉430000

出　　处：《现代电子技术》2017年第14期120-123,共4页Modern Electronics Technique

基　　金：国家自然科学基金项目(61300222)

摘　　要：为了提高4K超高清图像的成像质量,进行图像清晰化处理,提出一种基于改进神经网络的4K超高清图像清晰化技术。对采集的图像采用中值滤波降噪方法消除噪点,采用多尺度Retinex分解方法进行图像像素增强,结合改进神经网络方法进行模板特征匹配与信息融合,实现图像清晰化处理优化。仿真结果表明,采用该技术进行4K超高清图像清晰化处理,输出图像的信噪比较高,处理的时间开销和内存开销较小,具有优越性。In order to improve the imaging quality of4K ultrahigh-definition images and carry out sharpening processing ofthe images,a4K ultrahigh-definition image clearness technology based on the improved neural network is proposed.The medianfiltering denoising method is used to eliminate the noisy points of collected images.The multi-scale Retinex decomposition method is adopted to enhance image pixels.The template feature matching is performed in combination with the improved neural network method to deal with information fusion,so as to realize image sharpening processing optimization.The simulation resultsshow that the4K ultrahigh-definition images processed with this technology have high SNR,low expenditure of processing time,small memory overhead and a certain superiority.

关 键 词：神经网络 4K超高清图像 清晰化处理 降噪 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]