Quantitative evaluation for simple segmentation SVM in landscape image

Until now, designing a reliable image segmentation algorithm is still an open problem. Research related to this matter is still underway, but in one occasion we may be faced with the problem for selection image segmentation algorithms that will we use? To get the solution of this problem we need a good technical evaluation of image segmentation algorithms. With the technique, it is expected we can finally choose and use the right image segmentation algorithm. Pixel matching (Pm) an image segmentation algorithm evaluation techniques are common but considered less complete and does not support the refinement aspects evaluation. In this study we present two techniques for the evaluation of the segmentation algorithm: Local Consistency Error (LCE) and boundary matching. Furthermore both of techniques will use for evaluate segmentation algorithms based on Support Vector Machine (SVM) with a variety of simple features. In addition, as a comparison, k-mean will used as the base segmentation technique. From the experimental result showed that in general segmentation algorithm using SVM produces a better accuracy than k-means. The highest accuracy is obtained when the value is used as the SVM as classifier and Hue Saturate Value (HSV) as a feature. Sequentially evaluation value obtained was 90. 614% (Pm), 0.106 (LCE), and the highest value of precision and recall for matching boundary are 0.419 and 0.721 (when radius = 5 pixels).