Research On Key Technology Of Color Consistency Processing For Digtial Ortho Map Mosaicing

In recent years, the obtaining of high quality aerial images becomes more and more easy with the promotion of the sensor technology and the upgrading of the carring platform. The digital orthogonal map (DOM) which is made from the aerial images, as one of the4D productions in the surveying and mapping industry, plays an important role in the areas of cadastral surveying, urban planning, environmental survey, land administration, digital city, disaster monitoring, etc. Especially, the widespread use of a remote sensing platform--unmanned aerial vehicle (UAV), which is highly flexible, low dependency on weather, high-speed data acquisition and low cost, plays a timely key role and reduces the losses of property damage of the country and its people when facing emergent natural disasters, such as earthquake and flood.In the process of producing DOM, there are different degrees of discrepancies in the distribution of brightness and hue within and between images, caused by the acquisition time, illumination condition, camera angle and surface feature of the aerial images. These differences will directly affect the quality of the final digital orthogonal map. Particularly with the advent of the new remote sensing platforms, such as unmanned aerial vehicle and unmanned airship, the massive large scale images bring some new difficulties to the color consistency processing of aerial. images which cannot be solved perfectly by existing methods.This paper optimizes the processing of color consistency processing for Aerial Imageand improves the algorithm of the large scale aerial images processing. The main research contents and innovative results are as follows:(1) Solved the color cast problem on the edge of an aerial image after the Mask dodging processing. Halo phenomenon always appears on the edge of an aerial image after low-pass filtering in the frequency domain, which leads to a color cast problem in the dodging image. This paper analyzes the cause of the color cast phenomenon, and uses a background image optimizing method which is based on pixel replication to counteract the boundary effect of the gauss low-pass filtering in the frequency domain. The experiment shows that this method can effectively eliminate the halo that caused by the boundary effect, and avoid the color cast on the edge of the aerial image.(2) Proposed an optimized algorithm to solve the color cast problem caused by high contrast areas in the Mask dodging processing. Aerial images contain varied kinds of content. In some large scale images, large area of bright and dark spots that caused by some structures which have a strong specular reflection, will bring a significant influence to the dodging results. Especially in the two neighboring high contrast areas, such as high-brightness rooftop and large areas of the shadow in the wood, a color cast phenomenon always appears after dodging processing. This paper proposes a background image optimizing method based on expansion and engulfment algorithm. The experiments show that this algorithm can effectively eliminate the color cast phenomenon in the dodging results that caused by high contrast areas, and improve the quality of the dodging image.(3) Proposed a histogram smoothing algorithm to solve the discontinuous distribution problem of the gray level in the images after color consistency processing. With the consideration of the actual production, this paper chose a color consistency processing method based on a benchmark tone Wallis filtering. In the color consistency processing, when the variance of the benchmark image is higher than the variance of the image that need to be adjusted, a discontinuous distribution phenomenon of the gray level will appear after the Wallis filtering. Usually, the gray value of a high quality image should change continuously, and the gray level should also distribute continuously. This paper uses a histogram smoothing algorithm to avoid the discontinuous distribution of the gray level after color consistency processing. The experiments show that there is no discontinuous distribution phenomenon of the gray level by the method of this dissertation.(4) Proposed a tone transition method in the overlapping region after color consistency processing. There are still little differences among the tones in the overlapping region of the neighboring images after color consistency processing, which will influence the image quality in the image fusion. This research proposed a tone transition method in the overlapping region of the images to make the tone transition smoother in the overlapping region of the neighboring images. The experiments show that the method is effective, the tones of the two neighboring images in the overlapping regions are more similar with each other, and there is no obvious tone difference around the seam line after the image fusion.(5) Achieved the parallel processing of the color consistency processing for aerial image. As varied kinds of new remote sensing instruments are whipped into service, the obtaining of the high quality aerial images, which have large size, is more and more easy. We always face the insufficient memory problem and low computation efficiency when processing these large size images. Researching on the memory and time overhead saving has practical meanings to the actual production. This research used the parallel computing method based on Compute Unified Device Architecture (CUDA), analyzed and compared two typical fast computing algorithms of Fourier transform which cost a lot of time overhead in the dodging processing. After experiments, a parallel computing method based on CUDA was finally chosen to realize the color consistency processing for aerial image.