Kernel Clustering with Sigmoid-based Regularization

Introduction

This page contains source code for Kernel Clustering with Sigmoid-based Regularization (KCSR) for sequence segmentation. All the functions have been written and documented in Matlab format. Note that this implementation has no embedded C or C++ files (.mex). Therefore, it requires no any further installation or compilation. However, this convenience is achieved at an expense of a slight increase in practically running time.

Instructions

The package contains two folders and eight demo files:

• ./data: This folder contains four datasets, inlcuding synthetic data, Weizmann data, Google spoken digits data and ordered MNIST data.
• ./mfcc: This folder contains source code for computing del-frequency cepstral coefficents (MFCCs) from audio signals.
• ./demoSyn_FB.m: demo of KCSR on segmentation of synthetic sequence.
• ./demoSyn_SGA.m: demo of SKCSR on segmentation of synthetic sequence.
• ./demoWei_FB.m: demo of KCSR on segmentation of human action videos taken from Weizmann dataset.
• ./demoWei_SGA.m: demo of SKCSR on segmentation of human action videos taken from Weizmann dataset.
• ./demoGoo_FB.m: demo of KCSR on segmentation of Google spoken digits audio.
• ./demoGoo_SGA.m: demo of SKCSR on segmentation of Google spoken digits audio.
• ./demoMni_SGA.m: demo of SKCSR on segmentation of ordered MNIST digists sequence.
• ./demoMul_SGA.m: demo of MKCSR on segmentation of action video sequences of three subjects in Weizmann dataset.

The remaining files include: init_g.m, knGauss.m, knLin.m, KCSR_balanced_FB.m, KCSR_balanced_Multi_SGAm.m, KCSR_balanced_SGAm.m, KCSR_balanced_SGAo.m, sigmoid_mixture_cutoff.m, sigmoid_mixture.m, sigmoid.m, bestMap.m, hungarian.m and MutualInfo.m. They are all functional files that constitute the main implementation and evaluation of KCSR, SKCSR and MKCSR.

Notes

1. The default parameters in the demo files are adjusted on the datasets used in the paper. You may need to adjust the parameters when applying it on a new dataset.

2. We ultilized the code bestMap.m, hungarian.m and MutualInfo.m provided by Deng Cai (http://www.cad.zju.edu.cn/home/dengcai/Data/Clustering.html) which is publicly available. Please check the licence of it if you want to make use of this code.

Citations

Please cite the following paper if you use the codes:

1. Doan, Tung, and Atsuhiro Takasu. "Kernel Clustering With Sigmoid Regularization for Efficient Segmentation of Sequential Data." IEEE Access 10 (2022): 62848-62862.