摘要
研究了利用神经网络对序列黑白(灰度)图像进行着色的问题。针对以往基于人工或者半自动化技术的黑白图像着色技术效率低下、视觉效果较差的缺陷,提出了一种利用三层神经网络、无须人工干预的图像自动着色算法。首先将灰度图像分割成小块,通过对小块提取灰度特征、空间特征等作为神经网络的输入,训练得到一个回归神经网络。在着色过程中,可以利用该神经网络将图像中各像素由灰度空间投影到一个经过压缩的色彩空间,从而实现了图像的自动着色过程。实验结果显示本方法能有效地将灰度图像着色,并且由于使用了一个压缩的色彩空间,使得计算效率和着色效果都得到了有效的提高,能很好地逼近原始的真实图像。
This paper presented a neural network based colorization algorithm which could transform gray-scale images into color images.The traditional colorization methods have several shortages such as consuming time and performing is bad.First,the method segmented gray-scale image into smaller blocks.Then,based on a neural network which took the gray-scale and the location features of each block as its input,it could map the original gray-scale image into a compressed color space to reconstruct a new color image.This method had several advantages.First,the used features were simple.Second,because it used the compressed color space,the efficiency of the method was very high.The experimental results show that the result images are very close to the original color image,and this algorshm can provide better visual performance than others.
出处
《计算机应用研究》
CSCD
北大核心
2012年第4期1595-1597,共3页
Application Research of Computers
