Resources for the rectangular segmentation of an image (ML)

[Avatrin]

Hi

I see there are several articles about how CNN's are used to isolate and classify an object within an nxm rectangular region. While I know how to classify an image into one of p classes, I am not sure how to segment an image into rectangular regions which contain certain objects and, let's say, extract those regions.

I understand how to segment an image into non-rectangular regions by classifying each pixel by its and its neighbouring pixels values. However, I am not sure how to approach the problem of creating rectangular regions containing an object belonging to a class and extract that.

What are some good resources where I can learn to do this?

 

[Roberto Teso]

Google can be your best friend: rectangular segmentation.
For specific code examples you can learn to use OpenCV, a very good library for computer vision and machine learning with a lot of resources and a big community of users.

 

[Avatrin]

Well, Google wasn't of much help. The first page is full of papers for segmentation using rectangles, but it doesn't exactly give me an efficient method to extract a rectangular region containing an object; The methods are used for something different entirely.

I am just looking for a method which is smarter than the one that seems the most obvious: Finding the top, bottom, left- and rightmost pixels classified as belonging to class A and creating a region based on that (a misclassified pixel would completely ruin the segmentation + I cannot find multiple objects belonging to the same class in an image).

However, I made some progress; The term for the rectangular region I was looking for is a bounding box. So, my Google searches have improved. I guess I'll find something soon enough.