High frequency assessment from multiresolution analysis

Abstract

We propose a method for the assessment and visualization of high frequency regions of a multiresolution image. We combine both orientation tensor and multiresolution analysis to give a scalar descriptor of high frequency regions. High values of this scalar space indicate regions having coincident detail vectors in multiple scales of a wavelet decomposition. This is useful for finding edges, textures, collinear structures and salient regions for computer vision methods. The image is decomposed into several scales using the Discrete Wavelet Transform (DWT). The resulting detail spaces form vectors indicating intensity variations which are combined using orientation tensors. A high frequency scalar descriptor is then obtained from the resulting tensor for each original image pixel. Our results show that this descriptor indicates areas having relevant intensity variation in multiple scales. © 2009 Springer Berlin Heidelberg.