Hue Adjustmnt Method of Large-scale Image Database

This paper analyzes the reasons causing hue (or gray) differences of large-scale image database. Then based on the comparison of some general methods, the paper presents a new hue (or gray) adjustment and smoothness transition method according to the characters of image neighboring and correlative. Meanwhile, in order to reduce spatial transfer errors caused by adjusting large-scale images, an innovative multi-time adjustment algorithm based on quad tree is provided. 1. PREFACE Image data as a main carrier of spatial data is playing an increasingly important role in the field of geographic information System (GIS). It provides foundational data and information for constructing National Spatial Data Infrastructure (NSDI). Meanwhile digital orthophoto quadrangle (DOQ) is a core component of digital earth (DE) which was constructed on the base of spot image with 1-meter resolution. The construction of NSDI and DE imminently need a large-scale image database to efficiently manage the multiscale and multi-resource mass image data. At present, largescale seamless image database based on commercial database management systems (such as ORALCLE, SQL Server) has made significant theoretic contributions and successful applications both on the management of mass image data and on the matching and mosaic image in terms of spatial location information. However, there still exists serious non-equilibrium and deviance of the hue (or gray) in the large-scale image database due to different obtaining condition of original images. For example, different obtaining time of original image data will lead to different reflecting feature of objective and different sunshine conditions. Furthermore, even the obtaining time is simultaneous, the image data themselves will appear local deviance of hue (or gray) due to the influence of various factors such as terrain (the shade water area of mountain reflects and absorbs beam with different angles) etc. Seamed and mosaicked according to every single image’s spatial locations information the whole image will look like “a big clout” in large-scale (such as in a province or in the whole national area) seamless image database. Even though various hue (or gray) transition methods were performed on the seam line when images combined, the large range image still looks like “a big flower-cloth” for the hue (or gray) differences exist in every image. It is very difficult to maintain the consistent hue of a whole image in large-scale seamless image database. The existence of hue (or gray) deviance leads to the same object has different hue. And it badly affects objective recognition in full digitalize measure system and 3S (Global Position System (GPS), GIS, Remote Sensing (RS)). And also affects the whole appearance of output map of large range images. So how to guarantee equilibrium of large-scale images in practice becomes one of the key technologies of image mosaics in the construction of large-scale image database. 2. GENERAL METHOD OF IMAGE HUE BALANCE Most methods for image hue balance are based on the overlap areas of neighboring images. Various methods of smoothness transition are used to eliminate artificial spurious margins and mosaic seams existed in images. 2.1 Hue Balance Based on Image Histogram Gray-level histogram of images reflects their gray distribution characters. Under the ideal condition the gray-level histogram of the overlap area among neighboring images should be consistent. 2.1.1 Hue Adjustment Through Moving the Histogram Horizontally The average values of gray of left and right images are represented as: L g , R g . If we set the histogram of left image the reference and move the histogram of right image horizontally along the gray axis. Then we will have the formula: ( ) R L j i R j i R g g g g − + = ′ ) , ( ) , ( (1) An obvious advantage of this method is its minimal computation quantity. And it can achieve better result if the histograms of the overlap area of neighboring images have similar shapes. 2.1.2 Hue Adjustment Through Rejection of “First Order” Histogram The main character of this method is linear transform on graylevel according to gray distribution characteristics of Surface