Active surface model improvement by energy function optimization for 3D segmentation

作者：Azimifar Zohreh^*; Mohaddesi Mahsa

来源：Computers in Biology and Medicine, 2015, 59: 87-97.

DOI：10.1016/j.compbiomed.2015.01.014

• Active surface
• Wavelet edge detection
• Energy function
• 3D segmentation
• Local phase coherence
• Curvature

This paper proposes an optimized and efficient active surface model by improving the energy functions, searching method, neighborhood definition and resampling criterion. Extracting an accurate surface of the desired object from a number of 3D images using active surface and deformable models plays an important role in computer vision especially medical image processing. Different powerful segmentation algorithms have been suggested to address the limitations associated with the model initialization, poor convergence to surface concavities and slow convergence rate. This paper proposes a method to improve one of the strongest and recent segmentation algorithms, namely the Decoupled Active Surface (DAS) method. We consider a gradient of wavelet edge extracted image and local phase coherence as external energy to extract more information from images and we use curvature integral as internal energy to focus on high curvature region extraction. Similarly, we use resampling of points and a line search for point selection to improve the accuracy of the algorithm. We further employ an estimation of the desired object as an initialization for the active surface model. A number of tests and experiments have been done and the results show the improvements with regards to the extracted surface accuracy and computational time of the presented algorithm compared with the best and recent active surface models.

• 出版日期2015-4-1