METHOD FOR REDUCING IMAGE FUZZY DEGREE OF TDI-CCD CAMERA

摘要

The present invention belongs to the field of image processing, and particularly relates to the determination of an aerial remote sensing image fuzzy parameter and the elimination of aerial remote sensing image fuzziness based on a TDI-CCD camera. The method comprises the following specific steps: establishing an image coordinate system, reading an area array image, constructing a similarity matching criterion, conducting offset resolving to acquire homonymy points so as to obtain a digital image reducing the chattering influence. The method is relatively simple and precise in computing process, and good in processing effect.

主权项

1. A method for reducing blurs of TDI-CCD camera images, in which the target surface that supports the TDI-CCD camera is set to be n columns and m rows, characterized in that the steps are described as below:
S1. establish a coordinate system of the image: the direction of the number of pixels on the TDI-CCD camera area array, i.e., the direction of columns, is the Y axis, while the direction of the grades of the pixels on the TDI-CCD camera area, i.e. the direction of rows, is the X axis, the upper left corner of the image is the starting point of the coordinate system, wherein the coordinate for the first row and first column is (0,0); S2. read the area array images of the TDI-CCD camera outputted from each integration and carry out numbering of the images: the images are sequentially numbered as G1, G2, G3 . . . Gt . . . Gm; S3. carry out decomposition on the area array images: decompose each image read in S2 by rows, each area array image outputted from the integration grades is decomposed into m rows, each row having n pixels, each row forms a one-dimensional digital signal after decomposition, the signal length being n, the first N pixels are extracted from each one-dimensional digital signal, respectively forming one-dimensional digital signal series that is recorded as f(t), wherein 0≦t≦m, 0≦N≦n; S4. according to the one-dimensional digital signal series of S3, construct similarity matching rules for two one-dimensional signal series, including:
S41. select signal series f(i) as the benchmark series, carry out comparison between the signal series f(j) and the benchmark signal series, the neighborhood identicalness matching condition for the signal series is: ∥vij−vkl∥≦5, wherein ∥*∥ represents obtaining normal number, vij represents the disparity vector between any two pixels of the two rows of signal series, i represents the index of the pixels in the signal series f(i), j represents the index of the pixels in the signal series f(j), k is a neighborhood of i, l is a neighborhood of j;S42. calculate the initial matching probability according to said neighborhood identicalness matching condition of the signal series of S41:pij(0)=11+wij,wherein wijΣ|λ|≦k[g1(i+λ)=g2(j+λ)]2, pij(0) is the initial matching probability, g,(i+λ) represents the gray value of the one-dimensional signal series f(i) at the (i+λ)th pixel, g2(j+λ) represents the gray value of the one-dimensional signal series f(j) at the (j+λ)th pixel;
S43. according to relaxation method, establish the iteration formula of qij: pij˜(r)=Apij(r−1)+Bqij(r−1), whereinqij=∑k∑lpkl,r is the number of iterations, A and B are constants;
S44. according to the iteration formula of S43, obtain the post-normalization matching probabilitypij(r):pij(r)=pij∼(r)∑hpij∼(r),wherein, h represents every point that matches i;
S45. the converging diagonal series of p′ij can be known according to the matching probability obtained in S44, i.e., pij≈1, and other elements approach 0, wherein,pijr=[p0,0p0,1p0,2p0,3…p0,N-1p1,0p1,1p1,2p1,3…p1,N-1p2,0p2,1p2,2p2,3…p2,N-1p3,0p3,1p3,2p3,3…p3,N-1⋮⋮⋮⋮⋱⋮pN-1,0pN-1,1pN-1,2pN-1,3…pN-1,N-1]; S5. Conduct offset calculation according to the neighborhood matching calculation results of S4, obtain the tie points in the images, including:
S51. select the upper and lower neighborhoods corresponding to the benchmark signal series for cycle comparison, establish the corresponding relationship for the first pair of tie points g1(x1, y1) and g2(x2, y2) by comparing the benchmark signal series and the neighboring series in the next image;S52. determine the s known corresponding points of the images according to the corresponding relationship of similarity on the row dimension of the one-dimensional digital signal obtained in S51, in combination with bivariate quadratic polynomial, using least square method, obtain x1 and y1 by carrying out surface fitting of the data for the corresponding tie points:
x1=a00+a10x2+a01y2+a11x2y2+a20x22+a02y22 y1=b00+b10x2+b01y2+b11x2y2+b20x22+b02y22, S6. conduct spatial geometric correction on the images according to x1 and y1 obtained in S5, and use double direction linear interpolation to carry out pixel gray value reassignments; S7. carry out traversal operations on subsequent images relative to the benchmark image G1, traverse step S3 to step S6; S8. superimpose the one-dimensional signal series having the same instant field of view in the corrected images in the coordinate system to become the first row of digital image of the images with the oscillation effects removed, repeat step S3 to step S8, arrange each row of the superimposed image obtained with the reduced oscillation effects according to spatial and chronological order to form digital images with the oscillation effects reduced.