Is there a raster approach in machine learning?

Question

Algorithms

In such methods as a linear classifier, support vector machine, artificial neural networks, a search is made for a plane or surface that separates the points in the best way, while this surface is specified in a vector or parametric form. Are there methods in which at the output I would receive not a set of parameters, but a spatial raster image, each voxel of which would store a numerical value that determines whether all points inside this voxel belong to one class or another?
Below are our own attempts to implement this approach.
The following images were obtained by representing each point as an RBF, so that each point has an area of ​​influence that can be saved as a bitmap.
Tests on data models
Test with real data :