<![CDATA[The process of segmenting objects in compuer studies vision is very important because it is the basis for advanced image processing such as for object recognition. Automatic object segmentation is difficult to image with a complex background. Semi-automatic segmentation by region merging method is the correct method to perform segmentation of objects in a complex background. Semi-automatic method requiring interaction from the user in the form of markers that mark the object and the background to be segmented. Region merging method requires input in the form of low-level image segmentation results. The mean shift algorithm is ideal to use in low-level segmentation process, being able to split the image into many regions by keeping the borders of the object but the image has a weakness over the region segmentation, the resulting numbers are very overrated. The algorithm was able to overcome the problem normalizedd cuts over segmentation. Sequence region merging process with the help marker object and background markers from the user and maximal similarity based comparison of the color histogram of each region. From the results of experiments on 75 image dataset in getting the region merging method based on the input image of the mean shift + algroritma normalized cuts are very accurate in the object segment. This is evidenced by the value of bit error rate lower, reaching 0.09454 more accurate than the region merging algorithm simply mean shift are getting value for bit error rate of 0.20515. Improved accuracy obtained was 11%. Problems over segmentation also resolved, as evidenced by a decrease in the number of segmented region reached an average 69 %.]]>
