基于改进Keren配准方法的超分辨率算法

A Super-Resolution Algorithm Based on Modified Approach of Keren Registration Method

提出一种基于边缘检测和Keren配准方法的自适应归一化卷积超分辨率重建算法。为了进一步提高低分辨率序列图像间的配准精度,该算法将边缘检测与Keren配准算法相结合。首先利用Roberts算子对图像序列进行边缘检测,然后利用基于简化四参数仿射变换模型的Keren改进算法求出边缘图像间的平移和旋转参数。仿真实验结果表明即使在含有噪声及大角度旋转情况下,相比Keren改进算法该算法配准精度得到了显著提高;其中采用Roberts算子相比其他传统算子可获得更高的配准精度。最后采用自适应归一化卷积超分辨率融合算法进行超分辨率重建,真实混叠图像序列的实验表明,基于提出的这种配准方法的超分辨率重建图像获得了很好的视觉效果和更高的分辨能力,具有良好的应用价值。

An adaptive normalized convolution super-resolution algorithm based on edge detection and Keren registration method is proposed. To further improve registration precision of low-resolution image sequences, we combine Keren registration method with edge detection in the algorithm. Firstly, the edge feature of image sequences is extracted by Roberts operator. Then the registration parameters between edge feature images are obtained by Keren improvement registration method which is based on four parameters affine transformation model. The results of simulation experiment show that if the image edge detection is introduced before the Keren algorithm, the registration precision will be increased greatly even though there are noises and large rotation between images sequences. And a higher precision can be obtained by using Roberts operator compared with other traditional operators. Finally, the adaptive normalized convolution super-resolution fusion algorithm is used to reconstruct high-resolution image from low-resolution image sequences. Excellent reconstruction capability of the super-resolution algorithm based on the proposed registration method is demonstrated through an experimental real aliased image sequences.