Computer Vision News - March 2017

Computer Vision News Computer Vision News Research 27 Research In optimization functions of this sort there is always a trade-off between the local components and the global components. The local components influence segmentation locally; their disadvantage is that any local discontinuity can cause boundary leakage in the segmentation, because their local point-of-view can cause them to miss the wider context. Thus, the various improvements proposed over the years have tried to balance these components with global components that take the wider context into account. Method: In this article the authors propose the following improved level set equation: − ( [ ∙ + (1 − ] = 0 , ( , , = 0 = ∙ 0 ( , Let’s look at each component of the equation in turn: The initial condition of the function are set by 0 (x,y). This is a crucial step in level set segmentation method, greatly influencing the success of the segmentation. Poor initialization will result in failed segmentation, while mediocre initialization will make every subsequent step much more complex and therefore make for longer run-times, while a good initialization choice will simplify and streamline the entire process. In this article, the authors use Fuzzy c-means (FCM), one of the popular fuzzy logic algorithms, to determine the initial condition. The output of FCM is the membership function . which assigns every pixel (x,y) in the image to the k -th cluster with a probability between 0 and 1. This new formulation sets the following initial condition for the level set equation: That is, for every subset of clusters , we assign to the initial contour all pixels with a membership probability greater than . Where and are parameters of the algorithm. is the region competition component, which is the global factor of the segmentation equation. ( , | = 1,2,3 … = 2( ∈ > − 1 , ⊂ = ∈ − ∈ ∩( ∉