简体繁体 English

是否可以通过逆 gabor 变换“传输”图像纹理

[英]Is it possible to "transfer" image texture via inverse gabor transform

原文 2020-05-27 23:38:13 7 1 image-processing/ gabor-filter

Apologies if this is a naive or foolish question, but I am trying to learn a bit more about image processing techniques.如果这是一个幼稚或愚蠢的问题，我深表歉意，但我正在努力学习更多关于图像处理技术的知识。 I had an intuition about Gabor filters but can't seem to find an answer.我对 Gabor 过滤器有直觉，但似乎找不到答案。

If I calculate a bank of Gabor filters for a set of images and reduce them to N features that a machine learning algorithm has determined to be indicative of a specific texture, can these N features be applied to a novel image to "transfer" the texture to the novel image?如果我为一组图像计算一组 Gabor 滤波器并将它们减少为机器学习算法已确定指示特定纹理的 N 个特征，那么这些 N 个特征是否可以应用于新图像以“转移”纹理到小说形象？ Perhaps via an inverse Gabor transform?也许通过逆 Gabor 变换？ For example, if I have 10 Gabor filters that can accurate classify a texture as "brick", can these 10 filters be applied to a "wood" texture image (picture of a 2x4) to approximate the brick texture on the wood surface?例如，如果我有 10 个 Gabor 过滤器可以准确地将纹理分类为“砖”，那么这 10 个过滤器是否可以应用于“木材”纹理图像（2x4 的图片）以近似木材表面的砖纹理？

If possible this is possible, can it be easily implemented in Python?如果可能的话这是可能的，它可以很容易地在Python中实现吗？

1 个解决方案

As far as I understand, this is directly impossible.据我了解，这是直接不可能的。 "When working with Gabor filters, it is common to work with the magnitude response of each filter." “在使用 Gabor 滤波器时，通常要处理每个滤波器的幅度响应。” https://www.mathworks.com/help/images/texture-segmentation-using-gabor-filters.html That is, information is only about the magnitude of the signal, but there is no information about the phase. https://www.mathworks.com/help/images/texture-segmentation-using-gabor-filters.html也就是说，信息只是关于信号的幅度，但没有关于相位的信息。