Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/manuelblancovalentin/steerablepyramid

Steerable Pyramid Builder, Visualizer & Texture Synthesizer for MATLAB
https://github.com/manuelblancovalentin/steerablepyramid

Last synced: 25 days ago
JSON representation

Steerable Pyramid Builder, Visualizer & Texture Synthesizer for MATLAB

Awesome Lists containing this project

README 

        
# SteerablePyramid

Steerable Pyramid Builder, Visualizer & Texture Synthesizer for MATLAB



I created this code for my MSc. Thesis in Oil Reservoir Image Data 

 processing as I needed to understand how the workflow proposed by 

 Portilla et al worked. Even though Portilla's original code (see here: 

 http://www.cns.nyu.edu/~eero/steerpyr/) works great, it was very 

 difficult to fully comprehend each step in the different processes of 

 building the steerable pyramid, characterizing the textures and 

 synthesizing them, as almost no comments nor intelligible variable names 

 were used on the code. 

 I studied the code for several weeks and implemented all processes on my 

 own. I commented all steps and tried to include as many references to 

 the original paper as possible; so If anyone wants/needs to 

 understand how this code works, it'll be a lil' bit easier. 

 I have implemented ALL functions from scratch and I only use 1 of 

 Portilla's original code (expand function). I have also included another 

 function ('dispPyramid') which displays the steerable pyramid in a more 

 friendly way (giving the impression of a real pyramid), so that if you 

 need to display the pyramid in any paper or work, you can simply use 

 this function almost out the box. 

 I have tested my code and Portilla's original code using the same input 

 images and same initial conditions (starting white noise) and they 

 provide the EXACT same results (I checked this value per value). 

 This code can be used mainly for three purposes: 

  

  1) Build a steerable pyramid, given an input image, and a number of 

  desired scales and orientations. A steerable pyramid is a technique 

  that uses a recursive filtering workflow to obtain the different 

  attributes (texturally speaking) that an image may have at different 

  scales and orientations. This workflow basically consists on taking the 

  input image, filtering it with a highpass filter, a series of 

  different-oriented 2D band-pass filters and a low-pass filter. The 

  low-pass component (what's left-over) is then downscaled using a factor 

  of 2 and all the process is repeated again (High-pass + band-pass + 

  LP). By doing so we are extracting textural information about 

  orientation of the features in our image at different scales. 

  

  2) Extract Features that characterize the texture on the input image at 

  different scales and orientations (based on the steerable pyramid built 

  previously). These feature are those described on Portilla's paper too. 

  As shown on that paper they seem to be sufficient to characterize 

  several types of textures well (indeed they seem to do its job so well 

  that we can actually synthesize almost identical textures by using 

  them, as I will explain on the following entry). On the other side, 

  Portilla and Simoncelli asure they are universal parameters (which 

  roughly means that you can use it in a bunch of different applications 

  without having to implement much changes). 

  

  3) Texture Synthesis. As described in Portilla's paper, we can use the 

  textural features extracted from the steerable pyramid (used to 

  characterize it) to create a totally artificial and synthetic image 

  that will have the same textural characteristics as the original one. 

  On the other side, it is also possible to extrapolate texture, creating a 

  Mask that will preserve some part of the original texture, and synthesize the rest of it.