The Combination of Color and Depth Information for Importance Map

Abstract

In order to extract important map, a lot of image resizing algorithms only apply color map (such as gradient or salience information, etc.) to settle this question. On the analysis of depth map, this article thinks about depth information and combines with color map in order to put forward a combined important map. This new important map can identify the main body of source image accurately. Guided by this important map, image resizing results can not only protect the important area, but also obtain better processing effect. The image resizing algorithm can be segmented into two phases: the perception phase is the extraction for the importance map on the source image. The accuracy of important map affects the resizing effect directly. For importance map, most resizing algorithm only apply color information (such as gradient or salience information, etc.) to cope with the question. The importance map which is based on color map reflects the focus area of human eye, it is not quite accurately and its accuracy still needs to be improved further. In view of this situation, this paper thinks about depth information which is based on physiology (1) and psychology theory (2-4). Comparing with other automatic algorithm to calculate the importance map, quantitative results and qualitative analysis show that new algorithm may extract the importance map much more accurate. II. COMBINATION EXTRACTION ALGORITHM The joint of depth and color information map, the process of the combination extraction algorithm is shown on figure 1. The source information derives from the RGB-D camera. Combination extraction algorithm was divided into four parts: edge test, salience map detection, depth information generation and importance map's extraction. The edge test and salience detection belong to color information processing's category, depth information generation reside in depth information processing's scope, the extraction of importance map not only thinks about the color information in the source image (gradient or salience map), but also combines with the depth map. A. The Edge Test and Salience Detection With regard to edge and saliency test, we use sobel operator (5) to calculate the gradient information and adopt Itti model (6) to detect important map. New algorithm converts color image to grayscale image firstly, and then deals with grayscale image for edge test and saliency detection. Edge test is finished by applying the Sobel model which makes the calculated features keep the continuity on the source image's edge. Sobel model is a discrete difference operator and was applied to label brightness function's approximation on image. In the image's any pixels, we will calculate a corresponding gradient vector by using this operator. This operator contains two sets matrices respectively, they are the horizontal and vertical operator, which was used to deal with the plane convolution on image, we can get the brightness value of the difference approximation.