An Efficient Evolutionary Based Method For Image Segmentation

  • 2017-12-06 18:06:26
  • Roohollah Aslanzadeh, Kazem Qazanfari, Mohammad Rahmati
  • 0

Abstract

The goal of this paper is to present a new efficient image segmentationmethod based on evolutionary computation which is a model inspired from humanbehavior. Based on this model, a four layer process for image segmentation isproposed using the split/merge approach. In the first layer, an image is splitinto numerous regions using the watershed algorithm. In the second layer, aco-evolutionary process is applied to form centers of finals segments bymerging similar primary regions. In the third layer, a meta-heuristic processuses two operators to connect the residual regions to their correspondingdetermined centers. In the final layer, an evolutionary algorithm is used tocombine the resulted similar and neighbor regions. Different layers of thealgorithm are totally independent, therefore for certain applications aspecific layer can be changed without constraint of changing other layers. Someproperties of this algorithm like the flexibility of its method, the ability touse different feature vectors for segmentation (grayscale, color, texture,etc), the ability to control uniformity and the number of final segments usingfree parameters and also maintaining small regions, makes it possible to applythe algorithm to different applications. Moreover, the independence of eachregion from other regions in the second layer, and the independence of centersin the third layer, makes parallel implementation possible. As a result thealgorithm speed will increase. The presented algorithm was tested on a standarddataset (BSDS 300) of images, and the region boundaries were compared withdifferent people segmentation contours. Results show the efficiency of thealgorithm and its improvement to similar methods. As an instance, in 70% oftested images, results are better than ACT algorithm, besides in 100% of testedimages, we had better results in comparison with VSP algorithm.

 

Introduction (beta)

None

 

Conclusion (beta)

None