The existing technique for segmenting the target object the user wants from various images mainly uses two-dimensional features, resulting in various limitations due to the lack of three-dimensional information.This paper proposes a technique for segmenting objects robustly by combining two-dimensional and three-dimensional features from stereoscopic images that are received for effective clustering. In the proposed method, a stereo matching algorithm is used to acquire the depth informationexpressing the distance between the target and the camera. Then, the depth feature and the color feature are effectively clustered to detect the target object corresponding to the foreground.Through the experiment, it can be visually found that the proposed method accurately separates objects from the video with the help of depth information. However, it is sometimes difficult to extract depth information correctly because stereo matching is performed incorrectly in areas with similar color values or poor textures. To measure the performance of the suggested target segmentation approach, we defined the measure of Root Mean Square Error (RMSE). As a measure of dealing with general and specific aspects of image quality, RMSE is utilized to evaluate the difference between measured and actual values. In addition, it was confirmed quantitatively that the conventional method attempts to segment the target object using only twodimensional features, causing many errors, while the proposed method segments the target object more accurately than the conventional method by effectively clustering and using the distance information, which is a threedimensional feature without using only two-dimensional features.The suggestedobjectextractionapproach is expected to be useda lot in the practical areas of image analysis and pattern recognition, such as target object tracking andvideo surveillance.
Volume 11 | 07-Special Issue
Pages: 1755-1761