Object diameter measurement using Image Segmentation and Quantification

Image segmentation splits the original image into different non-overlapping parts to extract the desired region for various computer vision applications. Diverse methods exist to perform image segmentation, from traditional to AI; still, it is a challenging task in the image processing arena due to non-uniform intensities, poor contrast, shadow occlusion, irregular shape of an object, texture variations, noisy images, and different scalability. There is no global segmentation algorithm to address all these issues. Validation of the segmentation performance is also a difficult task, and it is mainly dependent on object detection rather than any standard metric. In this work, we performed image segmentation and quantification using the time-frequency domain to address segmentation issues like noisy images and multiscale resolution images for object detection. This work involves image segmentation using the wavelet thresholding method, then quantification by Haar wavelet transform, and finally, edge detection by Canny edge detection to obtain the median diameter of the object in the computer vision application. The proposed algorithm is flexible enough to adapt to different scaled images. The performance is validated with manually annotated data, and it exhibits 99% accuracy while locating the mean diameter of the desired object in the image.