angelounimi / palmnet

Source code for the 2019 IEEE TIFS paper "PalmNet: Gabor-PCA Convolutional Networks for Touchless Palmprint Recognition"

PalmNet

Matlab source code for the paper:

A. Genovese, V. Piuri, K. N. Plataniotis, and F. Scotti,
"PalmNet: Gabor-PCA Convolutional Networks for Touchless Palmprint Recognition",
IEEE Transactions on Information Forensics and Security, December 2019.
DOI: 10.1109/TIFS.2019.2911165

Paper:

https://ieeexplore.ieee.org/document/8691498

Project page:

http://iebil.di.unimi.it/palmnet/index.htm

Outline:

Citation:

@Article {tifs19,
    author = {A. Genovese and V. Piuri and K. N. Plataniotis and F. Scotti},
    title = {PalmNet: Gabor-PCA Convolutional Networks for touchless palmprint recognition},
    journal = {IEEE Transactions on Information Forensics and Security},
    volume = {14},
    number = {2},
    pages = {3160-3174},
    month = {December},
    year = {2019},
    note = {1556-6013}
}

Main files:

• launch_PalmNet.m: main file
• ./params/paramsPalmNet.m: parameter file

Required files:

• ./images: Database of images, with filenames in the format "NNNN_SSSS.ext", where NNNN is the 4-digit individual id, SSSS is the 4-digit sample id, and ext is the extension.
  For example: "0001_0001.bmp" is the first sample of the first individual. In the paper, left and right palms of the same person are considered as different individuals.

Part of the code uses the Matlab source code of the paper:

• T. Chan, K. Jia, S. Gao, J. Lu, Z. Zeng and Y. Ma,
  "PCANet: A Simple Deep Learning Baseline for Image Classification?,"
  in IEEE Transactions on Image Processing, vol. 24, no. 12, pp. 5017-5032, Dec. 2015.
  DOI: 10.1109/TIP.2015.2475625
  http://mx.nthu.edu.tw/~tsunghan/Source%20codes.html

the VLFeat library:

• A. Vedaldi and B. Fulkerson,
  "VLFeat: An Open and Portable Library of Computer Vision Algorithms", 2008,
  http://www.vlfeat.org

and the functions by Peter Kovesi:

• Peter Kovesi,
  "MATLAB and Octave Functions for Computer Vision and Image Processing",
  https://www.peterkovesi.com/matlabfns

The databases used in the paper can be obtained at:

• CASIA:
  http://www.cbsr.ia.ac.cn/english/Palmprint%20Databases.asp
• IITD:
  http://www4.comp.polyu.edu.hk/~csajaykr/IITD/Database_Palm.htm
• REST:
  http://www.regim.org/publications/databases/regim-sfax-tunisian-hand-database2016-rest2016/
• Tongji:
  http://sse.tongji.edu.cn/linzhang/cr3dpalm/cr3dpalm.htm

The segmentation algorithm can be found at:

https://github.com/AngeloUNIMI/PalmSeg

palmnet's Issues

Code description

thanks for your best contribution, I am very interested in your work. I am a beginner and tried to sort out your code, but I couldn't find the content of the network structure part. Can you help me point out the file part of the code about the palmnet network structure.Thanks

params in different database

In Tongji database,the EER is similar with paper . But in other database such as IITD,Casia. The EER is
much higher than paper about 3%. Do I need to tuning the parames?

How to select the filters?

Hello, I am a little confused about the paper. After getting Sm, how to extract 5 filters, could you please explain in detail?Thank you very much!

180-centers,why?

Hello, may I ask you a question? In the "searchGaborOrientation" code, on line 53, why do you subtract the centers from 180?Don't use 180 minus centers just the right bin?

Doubt regarding the selection of the most used wavelets (functions_Gabor/getMostUsedWavelets.m)

Hey thank you for this code and I appreciate taking your time to read this. I came across your paper about PalmNet in (IEEE) and it sounded very intriguing. So I was just looking through your code and I had a doubt in it. Please correct me if my understanding of this is wrong.

functions_Gabor/getMostUsedWavelets.m
In the above file, there is this snippet of code:

    %get information of corresponding wavelet
    currX = sortRes(ind, 4);
    currY = sortRes(ind, 5);
    currO = sortRes(ind, 6);
    
    %check if current wavelet at current position already
    %used
    if (tmpRes(currO).value(currY, currX) == 1)
        %increment counter
        ind = ind + 1;
        continue
    end %if (tmpRes(currY, currX, currO)

I assume it refers to this line from the paper:

Then, for each filter in Ga, we count the number of corresponding responses in sm, considering each filter at most once for each position (x, y) in the ROI.

But, doesn't sortRes contain a unique combination of O, X and Y? What I'm trying to say is in tmpRes, there'd be no value equal to 1 when that condition is being checked and that combination of O, X and Y would have never appeared before.

Please let me know if my understanding is wrong. Thank you for the code and once again I appreciate your time.

What does countW mean?

Hello, what does countW stand for in "getMostUsedWavelets"? Why is it capped at 10,000?