iFRAME (inhomogeneous Filters Random Field And Maximum Entropy)

Experiment 2: Wavelet Selection by Shared Matching Pursuit (Shared Sparse Coding)

Code and dataset

Wavelet selection by shared matching pursuit. Part (1) shows the training images. Part (2) shows the evolution of the sketch templates during the process of wavelet selection by shared matching pursuit. Each selected Gabor wavelet is illustrated by a bar, and each DoG by a circle. The template is the superposition of Gabor wavelets of 4 scales and DoG wavelets of 2 scales, where smaller scales appear darker. Part (3) shows the deformed sketch templates and the reconstructed images by the deformed sketch templates of each training image when the number of selected wavelets n = 1, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 150, 200, 250, 300, 350, and 400 pairs.

Case 1: Wolf

(1) Training images

(2) Sequence of sketch templates

(3) Sequence of deformed sketch templates and reconstructed images for each training image

training image 1:

training image 2:

training image 3:

training image 4:

training image 5:

Case 2: Cat

(1) Training images

(2) Sequence of sketch templates

(3) Sequence of deformed sketch templates and reconstructed images for each training image

training image 1:

training image 2:

training image 3:

training image 4:

training image 5:

Typical parameters:
image size is 100x100. The number of basis functions is 400 paris. The allowed range of perturbation in location is 2 pixels. The allowed range of perturbation in rotation is pi/16. GaborScaleList=[ 1.4, 1,0.7,0.5]. DoGScaleList =[18.90,13.36].