How to extract constant size roi image from an images in a dataset

In my dataset I need to extract the region of interest from an image to extract features from that image. But the problem is that the roi is different in size in terms of height and width. I want to extract the roi such that the quality of the roi does not degrade and all rois are of same size.

[Image Analyst]

Please answer these questions:

1. Is the ROI rectangular or not?
2. If rectangular, what should be row1, row2, column1, and column2 in each differently-shaped image? Would these change depending on the image size? If not rectangular, how would you determine its shape?
3. Why are your ROIs of different size?
4. Why do you think the quality of the ROI would degrade if the image is extracted from it or you're just processing withing the ROI? I don't see any reason why it would, as long as you don't resize it.

Please attach two sample images showing the ROIs for each image.

[shivasmic]

It is rectangular. And the size of roi is variable as the object inside the image is of different size. I need roi of constant size to extract the important features from the roi.

[shivasmic]

[Image Analyst]

Do you want to crop off the uniform gray frame and extract out only the gray scale image in the middle?

[shivasmic]

Actually ROI is been cropped using the coordinates from the bounding box function. But as can understand these coordinates vary from image to image. Yes I want to crop the grayscale roi using the same coordinates but such that the height and width remains same for each and every ROI image. Note: roi can be anywhere in the image not necessarily in the middle.

[Image Analyst]

Not sure what this means "Actually ROI is been cropped". Has it been cropped, or it still needs to be cropped?

Is the non-gray part of the image (i.e. the color image) the same size or not?

If they all have "the same coordinates" and you know the height and width, because they're constants ("the height and width remains same for each and every ROI" as you said), then why not just use imcrop()

croppedImage = imcrop(rgbImage, [col1, row1, width, height]);