Main topic:? How to achieve image stitching, find a good solution to find a long time. Author:? Wrcluomo (Fall) Grade:? ??? credit value:? 111 belongs to the forum:? VC / MFC graphics processing / algorithmic problems Points:? 200 Replies:? 21 Posted:? 2003-09-0410: 13: 59z ?????
There are two overlapping images, how to put these two images into a large image, remove the part overlapping each other. The hardest thing is how to find it overlapping each other? ??? Reply to: ASETT1 (ASETT1) () Reputation: 100? 2003-09-04 13: 17: 03Z? Score: 10 ??? From the left and right edge Search, calculate their respective grayscale covariance matrices, set the error range, and the two images can be fused within this range, and the remaining parts Copy can come. I understand this, let's go --_-
?? Top ?? Reply to: ASETT1 (ASETT1) () Reputation: 100? 2003-09-04 13: 18: 26Z? Score: 0 ??? Search from the edge of the left and right, calculate the respective grayscale covariance matrices, Set the error range, and the two images can be fused within this range, and the remaining part COPY is coming. I understand this, let's go --_-
?? TOP ?? Reply to: mattus_zhao (life ten eight or nine is not as good) () Reputation: 105? 2003-09-04 19: 45: 49Z? Score: 190 ??? First need to register images (Image Registering) 1 Method for using the phase correlation? Basic idea is to convert the image to the frequency domain, then find the highest point of the pulse, is the pan between the two images. The specific method is a bit complicated, you can view some articles about "correlation". 2 Select a control point between the two images and then determine the projection matrix between the images. For accurate projection matrices, a nonlinear optimization algorithm is required. The image stitching is then projected into a common space, and the image needs to be spliced. The easiest way is to integrate Alpha. Better and complex methods are multi-resolution image fusion, which requires the use of Gaussian pyramids and Laplas.
?? TOP ?? Reply to: wsljj_2002 (happy) () Reputation: 99? 2003-09-04 20: 14: 35Z? Score: 0 ??? Ask two cattle people, how is the different resolution Conversion? TOP ?? Reply to: WrCluomo () Reputation: 111? 2003-09-05 09: 58: 03Z? Score: 0 ??? Gray Method will not be too big, Fourier transform me tried Try is a bit slow, I see that the software of the people is very fast, and this method is definitely not used. I hope everyone will express more comments. This is a little late. ?? TOP ?? Reply to: WrCluomo () Reputation: 111? 2003-09-05 10: 19: 39z? Score: 0 ??? Gray method will not error, Fu Liye change I tried Try is a bit slow, I see that the software of the people is very fast, and this method is definitely not used. I hope everyone will express more comments. This is a little late.
?? Top ?? Reply to: mattus_zhao (life ten eight or nine is not as good) () Reputation: 105? 2003-09-05 16: 12: 52z? Score: 0 ??? Indeed, many commercial software's image splicing speed is very fast. I also made a stitching based on Fourier transform, but the speed is indeed slow. A better method should be: 1 global optimization. First determine the motion parameters between the images (this only approximately); 2 localization. Sampling pyramid method iteration refine. If such a method is used, the speed is very fast. However, the algorithm is high, I have not written it. Top ?? Reply to: mattus_zhao (life ten eight or nine is not as good) () Reputation: 105? 2003-09-05 16: 14: 31z? Score: 0 ??? Right, find the one of the overlapping There is a technology called "Best Seamline", but I don't know how to specific methods. ?? Top ?? Reply to: mattus_zhao (life ten eight nine nine nine nine nine is not as good) () reputation: 105? 2003-09-05 16: 16: 22Z? Score: 0 ??? I mainly refer to some IEEE and Siggraph top Article, if you are interested, I can write the name of the article. If you go to see, you will have a deep understanding of the image stitching. By the way, the image splicing has a very important role in the construction process of panoramic maps. ?? TOP ?? Reply to: WrCluomo () Reputation: 111? 2003-09-05 16: 39: 25Z? Score: 0 ??? Thank you life, find these articles there. I mean the grayscale method today and I can't find it. The following is the procedure I measure: The grayscale of the coincidence should be close to 0, but it can't see that the line is close. For (i = 0; i? {
??
?? for (j = 0; j ?? { ???? LPDATA [0] = LPDIBBITS [0] LLINEBYTES * (LHEIGHT - 1 -J) LWIDTH-3; ???? LPDATA [1] = LPDIBBITS [1] LLINEBYTES * (LHEIGHT - 1 -J) i; ???? ???? sss.format ("% D", ABS (* LPDATA [0] - * LPDATA [1])); ????? RgmyRge.SetItem (_variant_t ((long) (i 1)), _ variant_t (long) (j 1)), _ variant_t (sss)); ??} ?? ? 42? 227? 8? 219? 9? 192? 111 42? 223? 4? 233? 176? 133? 80 42? 223? 4? 233? 176? 133? 80 42? 223? 4? 233? 176? 133? 80 23? 233? 6? 49? 142? 174? 247 23? 233? 6? 49? 142? 174? 247 23? 233? 6? 49? 142? 174? 247 6? 227? 7? 109? 168? 229? 234 6? 227? 7? 109? 168? 229? 234 6? 227? 7? 109? 168? 229? 234 7? 220? 0? 97? 222? 6? 245 7? 220? 0? 97? 222? 6? 245 7? 220? 0? 97? 222? 6? 245 9? 238? 2? 249? 3? 10? 234 9? 238? 2? 249? 3? 10? 234 9? 238? 2? 249? 3? 10? 234 14? 21? 199? 236? 0? 10? 229 14? 21? 199? 236? 0? 10? 229 14? 21? 199? 236? 0? 10? 229 22? 9? 187? 17? 2? 15? 22622? 9? 187? 17? 2? 15? 226 22? 9? 187? 17? 2? 15? 226 45? 228? 232? 104? 2? 31? 235 Partial test data. If that is interested, I can send data from the test image to him. ? ? TOP ? ? Reply to: mattus_zhao (life ten eight or nine is not as good) () Reputation: 105? 2003-09-05 17: 39: 41Z? Score: 0 ? ? ? If you want to find research papers, the best place should be: citeseer.nj.nec.com The papers inside are very complete. Waiting for me to count, I have seen the name of the article, and I posted it. You can find this writing article by name. Of course, the article is just a reference, but it is indeed a lot. Foreign abroad It is very good, you can also see from commercial software. Many of the automatic splicing methods are Very accurate. ? ? TOP ? ? Reply: WrCluomo () reputation: 111? 2003-09-06 15: 25: 00Z? Score: 0 ? ? ? Give you extra points first, I open a post. First, thank you. ? ? TOP ? Reply to: Yellowwolf (you have given the name, so I register a wolf) () Reputation: 66? 2003-09-06 22: 43: 53z? Score: 0 ? ? ? I don't know what I use. You can find two drawings overlapping lines, but I don't know why there will be one more pixels (one pixel) No one helps me solve this problem, my code is behind. You can't see it. ? ? TOP ? ? Reply to: mattus_zhao (life ten eight nine nine nine is not as good) () Reputation: 105? 2003-09-07 08: 14: 20Z? Score: 0 ? ? ? You give me an email address through a short message, I will send it to you. But it is relatively large, you find a mailbox that can be collected. ? ? TOP ? ? Reply: Fanfyj (鹚 .NET) () Reputation: 100? 2003-09-07 08: 42: 18Z? Score: 0 ? ? ? GZ ? ? TOP ? ? Reply: WrCluomo () reputation: 111? 2003-09-08 17: 20: 30Z? Score: 0 ? ? ? my mail box: Power5318@sina.com Mattus_zhao, I posted you to you. ? ? TOP ? ? Reply: WrCluomo () reputation: 111? 2003-09-08 17: 21: 09z? Score: 0 ? ? ? This mailbox is small: send this: Wrcluomo@sohu.com ? ? TOP ? ? Reply to: quace (quace) () Reputation: 64? 2003-09-09 18: 36: 29Z? Score: 0 ? ? ? Quace@21cn.com ? ? TOP ? • Reply to: WrCluomo () Reputation: 111? 2003-09-11 08: 56: 09Z? Score: 0 ? ? ? UP ? ? TOP ? ? Reply to: mattus_zhao (life ten eight or nine is not as good) () Reputation: 105? 2003-09-11 09: 53: 12z? Score: 0 ? ? ? This time is very busy, go back to you at night. ? ? TOP ? ? Reply to: mattus_zhao (life ten eight or nine is not as good) () Reputation: 105? 2003-09-11 09: 56: 16Z? Score: 0? ? ? To wrcrluomo: Where is your post? How can I not see? ? ? TOP ? ? Reply: WrCluomo () reputation: 111? 2003-09-11 10: 32: 25Z? Score: 0 ? ? ? I am waiting for you for a few days, you don't go in, I have to disperse someone else. Also makes people misunderstand. see http://expert.9cbs.net/expert/topic/2230/2230647.xml?temp=.9627191 I will open it again. Fast forward. ? ? TOP ? ? This problem has been bombed, score record: ASETT1 (10), Mattus_zhao (190),????
