Abstract:
Automatic image segmentation is a very
important task for image analysis, object detection
and recognition tasks. In this research, automatic
image segmentation system is proposed which
includes three main approaches: preprocessing,
segmentation and post processing approach. The
preprocessing step estimates a better approximation
of gradient magnitudes by the modified 7x7
Laplacian of Gaussian (LoG) edge filter. In
segmentation step, marker controlled watershed
method (MCWS) is applied to solve oversegmentation problem. Finally, the segmented
regions are merged by using histogram similarity to
obtain the accurate segmented regions in an image.
This system is tested on two different kinds of
datasets: medical image dataset and color natural
image dataset. In this research, this system has also
achieved accuracy 93.01% for brain image, 76.72%
for color natural image. The running time of the
proposed system takes five times than MCWS method
for medical images due to region merging process for
many complex regions.
