laplacian matrix based on image data and local window size

Description

Input arguments:
imdata: MxNxd matrix. Image size is MxN, and feature number is d. Value range is within [0 255]
winsz: vec with 2-components showing size of local window for training local linear models. Values will be obliged to be odd.
mask: MxN matrix specifying scribbles, with 1 foreground, -1 background and 0 otherwise
lambda: para of the regularization in training local linear model

Output arguments:
L: (MxN)X(MxN) sparse matrix

reference :

Zheng, Yuanjie, and Chandra Kambhamettu. “Learning based digital matting” Computer Vision, 2009 IEEE 12th International Conference on. IEEE, 2009.

MATLAB code for rank of a matrix

MATLAB code for image correlation from time encoded Polaris SP data

https://en.wikipedia.org/wiki/Laplacian_matrix