所需积分/C币:10 2016-03-06 09:43:58 262KB PDF
收藏 收藏

1st International Symposium oil Systems and Control in Aerospace and Astronautics. Image mosaic is a very important step In big frame Image measurement,its precision has a great effeet on the measurement precision.twice matchim,gall make the mosaic suture precision beyond the pixel dimensions.And the pixeis’gray requirement rebuilding based on pixel position end pixel area distribufion ■lie has a great help to raise the measurement precision.
As mentioned above the offset in column direction are 4 attained by calculating the pixel grayscale-matching data The row offset is attained in the same way. After deal with the olumn and Tow offsets, we can get the precise relative location of two images with a higher matching precision. T DUAL WEIGHT IMAGE REBUILDING METHOD IN IMAGE SUTURE PROCESS 4 After the accurate position relation has been determined, the pixel grayscale of the new compositive image should be rebuilt. In the former image mosaic process, the pixel positions don' t need to be rebuilt since there is no offset in the Fig3 Grayscale alteration curve of corresponding column in onp/nal sub-pixel dimension, so the pixel position of image is still at the original location, only the overlapping area need to rebuilt. Consider that tracking the curve peaks directly requires When images have been finished the sub-pixel dimension complex curvefitting, it does not meet the need of fast level matching, the distribution of pixel position will not calculation, so we must choose a proper algorithm. Otherwise strictly align in the same position as the original images, as we do not concern the peaks themselves and their positions, in the compositive image frame should be repartition and the final and biggest focus is the exact offsets between peaks. So we only need to track the alteration trend of the curve and get the results using least square curvefitting method in the increasing section and decreasing section respectively. As shown n Fig 4, the new offsets are attained after the least square curvefitting method, which has the same precision of original offsets while calculation is greatly reduced Fig 5 Pixel distribution in compositive image In the figure, blue and red area show pixel positions in 42 original image. Set the image frame in blue as reference, pixels in green should be repartition, the image in overlapping area should be rebuilt according to the pixel location, and the pixel grayscale in green area should be recalculate according to pixel arca distribution and pixel 么 grayscale weight A. Image grayscale weighting based on pixel position The Image grayscale weight based on pixel position Fig 4 Sketch map of curvefitting in corresponding colum represents the relation between pixel's grayscale and the By tracking the grayscales value, we divide them into grayscale in corresponding position of two original image, increasing section and decreasing section, the consecutive the value of the grayscale at one pixel point is weighted by the oints are set to(r, Gi)and(X,G),i=,2,m distance between the pixel position to the edge of the overlapping area. As shown in the Fig. 5, D is the width of two j=1, 2,, n, while G=a+bX,, fit this two group of frame's overlapping area, dl is the distance of the pixel point data using least square method and attain(3. b, ), (a, b, ),. to left edge of overlapping area, d is the distance of the pixel point to right edge of overlapping area, assume the grayscale he junction of two adjacent lines is the peak. Work out the value of two original frame are Gl and G2, the grayscale equation: a+bX=a,+b,-X, we can get the position of value after being weighted should be the peak, Do the same with the peaks left, then we can Gxd1+g.xd2 calculate the,4、43、4… Usually we set the D average value a as the final parameter for fixing image position B. Image grayscale weighting based on pixel area distribution Image grayscale weighting based on pixel area distribution represents the relation between one pixel's grayscale in compositive image rebuilding area and the grayscale value at the same position in original images. The grayscale is weighted by the pixel area distribution in original images. As shown in Fig. 6, the frame in red is original pixel area, and the green area is pixel area in rebuilt image, Gl, G2, G3, G4 are the grayscales of each pixel in original image, S1, S2, S3, $4 are area in rebuilted image. S is the area of one pixel. The image grayscale weighting based on pixel area distribution chould be GxS1+G2XS2+G3×S3+G4×S4 G1 G2 SitZ G3)S3 G4 Fig. 7 Ongindl images need be matched in left-nght direction Fig. 6 The pixel distribution map and the regions in red and green frame sign the or Image area V. VALIDATION BASED ON EMULATIONAL EXPERIMENT Input images Fig 7 shows two images taken at the same place with almost 40 percent superposition area. Since the images had Choose sensitive area template been finished in twice shoot so there is nonzero offset between them, and they are fit for validating the image mosaic method. These two images had been shot at the same Image matching at the pixei dimension levet time almost, so the lighteness are almost the same too. For validating the adaptability of the root of grayscale average Image matching at the sub-pixel dimension level in column direction difference in difference matrix in sensitive area template matching process, we add 5 grayscales to every pixel's grayscale value Image matching at the sub-pixel dimension The mosaic emulational experiment uses Matlab 6.0 to level in row direction compile functions for the image matching and rebuilding. The process of image mosaic is shown in Fig 8, and choose 20X20 rectangle area in the overlapping region as sensitive Rebuid the pixel area distribution of area template, as the blue rectangle area shown in Fig 9. At mosaic image the best matching position, the root of grayscale average difference in difference matrix is,=0.6703. In the Rebuild the grayscale of every pixel sub-pixel dimension level matching process, the offsets attained after the least square curvefitting are 0.3 pixel Output images dimension unit in row direction and 0.45 pixel dimension unit in column direction method, the mosaic image is shown in Fig 8 Flow chart of image mosaIc Fig9. 206 Fig g The mosaic image VI. CONCLUSION After twice image matching, the precision of image mosaic has been enhanced obviously. At the same time, the dual weighting operation in image rebuilding makes the mosaic image more higher precision. According to the flow chart to finish the image mosaic process, it could fit the high-precision image inspecting very well at image precision aspect, and this method has a very important significance to get rid of the positional error of the platform and to enhance the precision of inspecting system. REFERENCES [U Junjie Wang, Jiamao Liu, Yunfa Hu, "Image Mosaic technics Computer Science 2003. [2] Hanguo Cul, Jun Chen, Maochun Cao, "Research of image MosaIc and Navigation algonithm Based on Panoramic Cylinder Image", Joumal of Naval University of Engineering, 2004 2 [3] RBrunelll, T PoggIo"Template Matching: Matched spatial filters and ond", Patten Recongnition, 1997,30 [4] LI Zhong, XIaofeng Hu, Overlapping Image MosaIc Algorithm Journal of image and GraphIcs 19985 [5] Hongiie Xle Automatic image registration based on FFT algonthmand IDL/ ENVI, iniematona) Conference on Remote Sensing and GIS/ GPS2ICORG, 2001 [6] Yuanping Zhu Limn Xra, A Robot Template Matching Algorithm fo Automatc Image Stitching", Computer Engineenng and Applications 2003(31 High-precision Image Mosaic Method in Big Frame 旧 万数据 WANFANG DATA文献链接 Measurement 引用本文格式: High-precision Image Mosaic Method in Big Frame Measurement[会议论文]2006

试读 6P High-precision+Image+Mosaic+Method+in+Big+Frame+Measurement
限时抽奖 低至0.43元/次
身份认证后 购VIP低至7折
关注 私信
High-precision+Image+Mosaic+Method+in+Big+Frame+Measurement 10积分/C币 立即下载

试读结束, 可继续阅读

10积分/C币 立即下载