Browsing by Author "Wibisono, Jan Kristanto"

Now showing 1 - 2 of 2

Fusion of Color and Depth Information for Image Segmentation
(2016)
Wibisono, Jan Kristanto
;
Hang, Hsueh-Ming
;
交大名義發表
;
National Chiao Tung University
The goal of this research is to fuse color and depth information to generate good image segmentation. The image segmentation topic has been studied for several decades. But only recently the use of depth data becomes popular due to the wide spread of affordable and accessible depth cameras such as Microsoft Kinect. The availability of depth information opens up new opportunities for image segmentation. Many methods have developed on color image segmentation over the years. Only recently, several papers are published on image segmentation using both the depth information and the color information. In this research, we focus on how to combine the depth and color information to improve the state of art color image segmentation methods. We adopt a few existing schemes and fuse their outputs to produce the final results. We exploit the planar information to improve the color segmentation. The result is quite satisfactory on both human perception and objective measures.
RGBD IMAGE SEGMENTATION USING DEEP EDGE
(2017-01-01)
Wibisono, Jan Kristanto
;
Hang, Hsueh-Ming
;
交大名義發表
;
National Chiao Tung University
In the past a few decades, many schemes have been proposed for segmenting a color image into meaningful regions. However, the newly availability of depth data provides opportunities to explore and improve the image segmentation performance. In addition, the new image processing tools based on deep learning technology are aggressively developed recently. This paper proposes a method of combining color and depth data to segment an image. As an initial stage, we partition a color image into regions using the DeepEdge tool, an image edge detection scheme developed based on the CNN (Convolutional Neural Net) technique. Then, we use the RANSAC tool to identify and merge regions with similar planar geometry (based on the depth information). At the final stage, guided by the DeepEdge information, a region merging method is employed to fine-tune the merged regions based on the color and depth similarity. Comparing to our previous results, the DeepEdge method together with the depth information helps in improving the segmentation result in most cases.