Search code examples
pythonopencvimage-processingimage-segmentationedge-detection

Which Image Quality Metric should I use in this case (Binary Segmented Images)?


I have some ground truth Image that I want to compare with my automatic segmentation code. I need some help to choose the correct Image Quality Metric, I've heard about ssmi,psnr, but knowing that I'm working with binary images, I think I need a different kind of Metric. Do you guys have any idea?

Ground Truth Image:
Ground truth Image

My segmentation code:
SegmentedAutomaticly


Solution

  • Since you are comparing segmentation masks, PSNR and SSIM are not very good measures.

    You should look at Intersection over Union (IoU) of the output mask and the ground truth. Or the distance of the boundary pixels from the ground truth.

    Berkeley segmentation dataset focuses on the task of image segmentation and boundary detection, which seems closely related to your task. You can check out their methods for evaluating segmentation results.