Comparative analysis of standard image segmentation methods

Image segmentation is key process in image analysis which is based on separating "objects of interest" from the rest of the image. This represents the first step in obtaining meaningful information from the images, this enabling further analysis of the identified components. In this paper a comparative analysis of standard methods for image segmentation, namely the K-means algorithm, the Watershed algorithm and the Region Growing algorithm, was performed. For this research were used real-life image from BSD500 image database, specifically 200 Training Images – Berkeley Segmentation Dataset. After the segmentation of the image, the results were processed first with the mathematical indicators PSNR and then with the metric for assessing the accuracy of the model with F1 Score metrics. The obtained results show that for the PSNR metric, the Watershed algorithm gives the lowest results for the PSNR metric, with an average value for PSNR of 6.67dB, in contrast to Region Growing, which stands out significantly in this case with 9.73dB, while the F1 Score metric singles out K-means as the best in performance. With the best response (26.82% faster response than Watershed and 109.01% faster than Region Growing algorithm) but week precision.