Robust color multi-focus image fusion using quaternion sparse representation and spatial information

Most sparse representation (SR)-based fusion methods handle color channels separately, which easily causes hue distortion and color saturation reduction in fused images. To address these issues, we propose a novel color multi-focus image fusion method based on quaternion sparse representation (QSR). QSR employs the quaternion matrix to model color images in a holistic way that can fully exploit high correlations among color channels and preserve the inherent color structures of source images in reconstruction results. We first learn a clear quaternion dictionary on a high-quality image set and a blurry quaternion dictionary on a Gaussian-blurred version of the same set. To accurately estimate the focus information of each image patch, the salience and sparsity features are computed by collaboratively employing the sparse coefficients of the current patch and its neighboring patches deduced from the QSR model under the two learned dictionaries. Then, an activity measurement of each patch is defined by exploiting the two features. Finally, a maximum fusion rule is adopted to get the composited sparse coefficients. The experimental results show that our method successfully avoids color distortion in the fused images and performs qualitatively and quantitatively better than some recent SR-based fusion methods.