How to regularize more strongly (working with sparse images)?

[emmenlau]

Thanks for BaSiC which generally works awesome. However we sometimes find issues when processing fluorescence images from plates with sparse cells. I'm under the impression that there are two potential pitfalls for BaSiC:

• If the cells are sparse and small, the imresize() may not capture them at all? It will use nearest neighbor, as far as I could see, so it may just happen to not grasp the relevant foreground pixels?
• The model from BaSiC is sometimes quite irregular, with strong local variations, i.e. in cases of dirt or sparse cells. I think it would help to regularize more strongly, but I do not know how to sensibly do that in the code. Actually I am often under the impression that a resolution of 32x32 instead of 128x128 would be sufficient, but I do not feel comfortable to change such integral settings of the method. Do you have recommendations for regularization of the model?