一种基于K-均值分类稀疏表示的灰度图像颜色重建方法

Method of color reconstruction of greyscale images based on sparse representation of K-means classification

充分利用参考图像与待处理灰度图像的关联关系,运用稀疏表示理论和字典学习的方法,提出一种基于K-均值分类和残差补偿的稀疏表示的方法来对灰度图像进行颜色重建。首先根据K-均值算法将参考图像分成K类,利用K阶奇异值分解(K-SVD)算法训练各类的亮度—特征—颜色的联合字典;其次,根据最小形心距离将待处理灰度图像自适应地分成K类,利用其亮度和特征信息根据正交匹配追踪(OMP)算法得到各类的稀疏系数;然后利用各类的字典和稀疏系数重建初始的彩色图像;最后用残差补偿对重建结果进行修正。实验结果表明,该算法相比于经典算法及其他改进算法对灰度图像进行颜色重建时取得了更好的效果,重建的图像看起来更自然、平滑,并且在客观评价标准方面也优于对比算法。

Considering the relationship between reference color image and the greyscale image, this paper proposed a method of color reconstruction of greyscale images based on sparse representation of K-means classification and residual compensation applying the method of sparse representation and dictionary learning. Firstly, it classified the reference image into K clusters by the K-means algorithm and trained the dictionary of the combination of luminance, feature and color information of each cluster by the K singular value decomposition （K-SVD） algorithm. Secondly, it sought the cluster of the original greyscale image adaptively based on the minimum between the value of each patch and the centroid of each cluster, and calculated the coefficients of each cluster by the orthogonal matching pursuit（0MP） algorithm according to t.he combination of luminance, feature information. Then, it reconstructed the primal color image of each cluster using the corresponding dictionary and sparse coefficient. Finally, it used the residual compensation to fix the reconstruction result. Experimental results demonstrate that this approach outperforms other approaches compared with the classical algorithm and the improved algorithm based on sparse representation in both subjective and objective criteria.