July 02 Reviewed: August 2022
FIGURE 3. An example of the location of points according to the first scheme. FIGURE 4
Download 0.5 Mb. Pdf ko'rish
|
50 13 34 1-6
FIGURE 3. An example of the location of
points according to the first scheme. FIGURE 4. The example of the location of points according to the second scheme. FIGURE 5. An example of the location of points according to the third scheme. FIGURE 6. Algorithm for finding area. • for the second group, the location of reference points along the perimeter of the corrected area is selected (scheme 2, the ‘perimeter’ method), respectively, the number of points is equal to the side of the square multiplied by four (Fig.4); i from 0 to w, step 1 d from 0 to 4, step 1 t = t + 2 End Begin d_min > d_test sign // Signature of the corrected area L // Width of the corrected area w // Search width d_min // Minimum distance between colors bmp // Image to adjust old_x, old_y // Old coordinates test_x, test_y // Coordinates to be checked new_x, new_y // New coordinates k from 0 to L-1, step 1 t = 0 test_x = old_x + i test_y = old_y + j j from -i to i, step 1 d = 0 test_x = old_x - i test_y = old_y + j d = 1 test_x = old_x + j test_y = old_y + i d = 2 test_x = old_x + j test_y = old_y - i d = 3 d_test = (√(sign t .R-bmp test_x+k,test_y+k .R) 2 + (sign t .G-bmp test_x+k,test_y+k .G) 2 + (sign t .B-bmp test_x+k,test_y+k .B) 2 ) + √ (sign t+1 .R-bmp test_x+L-k,test_y+k .R) 2 + (sign t+1 .G-bmp test_x+L-k,test_y+k .G) 2 + (sign t+1 .B-bmp test_x+L-k,test_y+k .B) 2 )/2 d_min = d_test; new_x = test_x new_y = test_y 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 • for the third group, the location of the reference points on the main and additional diagonals of the corrected area is selected (scheme 3, the ‘diagonal’ method), respectively, the number of points is equal to the side of the square multiplied by two (Fig.5). The fact is that the convolution filtering algorithm includes a multiplication operation, as a result of which the number of code bits that represent the samples of the filtered image turns out to be equal to the sum of the number of code bits used to represent the original image and the impulse response. Since both the original image and the impulse response are usually represented by eight-digit numbers, the result of filtering is an image that requires a sixteen-digit code to accurately represent the intensities. In order to switch to the previous type of recording of the filtered image, it is necessary to bring it into an eight-digit representation, i.e. to round off, which causes rounding noise. The order of each group of experiments was the same and consisted of the following steps: 1. Open one main image and five additional images, these images are six consecutive frames of a certain video sequence. 2. Select a 10 by 10 pixel area on the main image. 3. Sequentially search for the selected area on each of the additional images. 4. After the end of the search on the next image, mark on it the actual position of the selected area, found manually. 5. Estimate the search error on each of the additional images as the length of the straight line connecting the upper left corners of the found areas. 6. Estimate the search time on each of the additional images in seconds. 7. Enter the evaluation results in the table. 8. Repeat steps 1-7 for the next two groups of experiments. The results of the experiments are shown in Tables 1 and 2. The experiments were carried out on a personal computer with the following configuration: Processor: Pentium 4, 3.0 GHz; RAM: 2048 Mb, DDR2-800; Video card: built-in, Intel 82945G Express; Operating System: Windows XP SP3. The following conclusions can be drawn from the above results: - although the ‘4 points’ method works much faster than the others and does not depend on the size of the selected area, it also gives the greatest search error, so it is not applicable in this task; - the ‘perimeter’ and ‘diagonal’ methods give approximately the same and fairly low error, but the second method is much faster than the first, so it was the ‘diagonal’ method that was used to implement the search in the program. After the end of the search, new coordinates of the correction area will be obtained, according to which the control coefficients obtained at the control stage will be applied. The use of predefined coefficients significantly speeds up the processing of the corrected area, since at the control stage, when working with the main image, most of the time is lost on their search. The processing time decreases in proportion to the size of the corrected area, on average up to 20 - 2000 times. The process of applying coefficients is no different from the control process, except that instead of a new search for control coefficients, previously found values are simply substituted. A brief description of the algorithm (the functional blocks of the algorithm are indicated in parentheses): the algorithm begins (1) by opening a loop from 0 to the search width (set from the interface) with a step of 1 (2), then the loop opens from up to 4 (sets the tested side of the testing square) in increments of 1 (3), then a cycle opens in width in increments of 1 to pass along the side of the testing square (4), in which it is checked which side is selected for testing (5,6,7,8), depending on which the coordinates of the upper left corner of the testing square (9,10,11,12) are set. Next, we set the index of movement by the signature (13) and open a loop along the width of the corrected area (14), in which we calculate the difference between the current points of the test square and the original area (15) and then compare it with the minimum of the previously found differences (16): if it is less, then the value of the minimum difference is updated and the coordinates at which this happened (17) are remembered, then we increase the index of movement by the signature (18). The algorithm is completed (19). Methods have been developed for solving game pursuit problems with second-order difference equations. When solving this problem, an explicit formula is used, which is described using the Chebyshev polynomial of the second kind. A two-dimensional raster model of the image of scenes and their constituent objects has been developed. Models of entropy encoding of television images in video information systems have been developed. Brightness control systems using Chebyshev matrix polynomials have been developed. An algorithm for searching for reference brightness correction coefficients and an algorithm for calculating Chebyshev matrix polynomials are given. |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling