How can I add custom pixel values in a 4th layer matrix of an RGB image?

Hi I am quite new to matlab so excuse my amateur way of expressing the problem. Let's say I have truecolor "initial_image" MxNx3 and I want to create a "4d_image" identical to the "initial_image" which will be a MxNx4 or just a new matrix MxN which will descrive the initial_image objects in a pixel level.

First, I want to be able to manually choose a pixel region in the "4d_image" and then for that certain region I want to manually add a new value to each pixel (without any computer vision or image processing algorithm). For example all the pixels of the "4d_image" which are in the rectangle [46:100 3456:3890] that I manually chose are of type "door" (that ofc will correspond to a numeric value let's say from 0 to 255 so that it won't be a string).

So the output image would be an MxNx4 image in which the 4th layer matrix will have the information about the different objects included in the image in a pixel level. Or it will be a new 2-D matrix Q(M,N) that will describe each pixel of the initial image in the right i,j position.

Do you have any advice on how to start? I already checked some functions like "imfreehand", "impixel", "impixelinfo", etc but I am getting lost on how to extract those features and use them to create the 4th matrix and thus the 4d_image. What subjects or kewords should I look for?

[jonas]

Sounds like you are describing a 2d image represented by a weirdo 4-layered 3d array. How do you want to process this array once you have attached this 4th layer to the original RGB triplets?

[Orfeas Skentzos]

Let me rephrase my question then. What is the best way to inspect an RGB image? If I want to inspect it pixel by pixel and add new values in each pixel that describe the object showing? Which functions are useful for choosing manually certain regions and then adding values to those pixels? I just need some headstart tips in order to look for the right coding because I keep getting lost.

[jonas]

Just use your mxn array to make a mask and apply said mask (logical indexing) to your image.

If you provide an example I (and likely others) will be happy to assist

[Guillaume]

What you describe is called labeling (e.g. this group of pixel has label 1, this group of pixel has label 2, etc.). Typically, the labels are stored in a separate 2D matrix, not as a 4th layer.

What I haven't understood at all is what criteria you want to use for labelling your pixels. A typically labelling method is to find connected pixels in a binary image. You may also be after some segmentation method, I'm really not sure.

[Orfeas Skentzos]

For instance I am having an MxNx3 image of a building facade and I want to label the regions with values from 0-255 lets say. Each number will correspond to an object (i.e 1--> door 2-->window 234-->human_noise etc).

So the input will be an image MxNx3 and the output will be something like what you describe, a 2D matrix with MxN dimensions (same as the input image) in which all regions that are doors will have the label 1 all regions that are windows will have the label 2 etc. But I specifically want those labels in their right (i,j) position which will correspond to the same (i,j) position of the input image.

I don't know is this similar to a color map? Is there a way to interact with the image and do that process manually? I am still in very early stages so I cannot provide a good example. Instead, that's what I am looking for in order to approach the problem correctly.

[Image Analyst]

See my second answer below.

[Guillaume]

But I specifically want those labels in their right (i,j) position which will correspond to the same (i,j) position of the input image. I don't know is this similar to a color map?

This is exactly the description of a labeling matrix and is commonly used in image processing. Matlab has several functions to work and generate labeling matrices, see for example label2rgb. It is not a colour map.

_ Is there a way to interact with the image and do that process manually?_

Note that you want two different things: a) segment the image in different blobs. b) label the blobs.

For the segmentation it would be better would be to come up with a way to do it automatically. In any case, the segmentation link I gave earlier gives plenty of methods to segment the image. The second link on that page points to the Image Segmenter App which lets you manually select regions.

[Orfeas Skentzos]

Thank you Guillaume. That's really helpful! I will check the links.

Answer 1

[Image Analyst]

You can add another plane to your 3 plane RGB image like this:

image3d = cat(rgbImage, image2d);

Note: adding a 4th plane does not make the image a 4-D image like you hinted - it's still a 3-D image just with 4 channels/slices/planes.

[Orfeas Skentzos]

Yeah maybe the naming of the new image output was misleading. Thanks for the tip. I will give it a try. But first I have to figure out how I will manually do the labeling so then I can insert this information to the new plane as you indicated.

[Guillaume]

As commented in the question, it is much easier to keep the label matrix separate from the image matrix as matlab image processing functions won't understand what that fourth plane represent.

Answer 2

[Image Analyst]

See my classification demo where you can draw regions. You could then use those regions to build up a labeled image, instead of doing color segmentation like the demo does.

[Orfeas Skentzos]

Thank you. I'll check it and try to modify it.