needing help in algorithm
-
if it is a bmp, then i think you could go though each pixel and test the pixel around like this (i'm not an expert of image computing, so i could be wrong) :
for each vertical line -> 'y' coordinate
for each pixel in the line -> 'x' coordinate
if pixel color is red then
if pixel directly on top is red and
pixel directly at left is red and
pixel directly at right is red and
pixed directly at bottom is red then
you found the plus symbol, which coordinates are (x; y)
end if
end if
end for
end for
TOXCCT >>> GEII power
[VisualCalc 3.0 updated ][Flags Beginner's Guide new! ]
I know this way but it is a lot of noise in image and the plus isn't like this : ------------o-------- -----------ooo------- ------------o-------- it is like this : -----------------------o----------- -----------------------o----------- -----------------------o----------- ---------------o o o o O --- o o o o ---------------------o------------ ----------------------o----------- ----------------------o----------- ----------------------o----------- and the most important is that it may rotate . and i dont want to rotate it in a ( for ) I want it to be a speedy algorithm . and i also want to use the long tail of plus , I mean it isn't one pixel . thanks -- modified at 10:19 Thursday 11th May, 2006
-
I know this way but it is a lot of noise in image and the plus isn't like this : ------------o-------- -----------ooo------- ------------o-------- it is like this : -----------------------o----------- -----------------------o----------- -----------------------o----------- ---------------o o o o O --- o o o o ---------------------o------------ ----------------------o----------- ----------------------o----------- ----------------------o----------- and the most important is that it may rotate . and i dont want to rotate it in a ( for ) I want it to be a speedy algorithm . and i also want to use the long tail of plus , I mean it isn't one pixel . thanks -- modified at 10:19 Thursday 11th May, 2006
edit your post and use the <pre></pre> tags to preserve spaces...
TOXCCT >>> GEII power
[VisualCalc 3.0 updated ][Flags Beginner's Guide new! ]
-
hello I have a image , and a red plus is in it , and Iwant to find it's X and Y. and I want a algorithm . thanks a lot
I want, I want, I want. Since this site is not about doing the job for you, but helping you with problems, I will point you in a good direction and wait to see what effort you make. Search code project for the following types of article image processing and pattern recognition Some thoughts to consider are: 1) Are you looking for a specific size red plus or any two crossing lines of any thinkness and any length? It becomes easier if you know the exact size, thinkness of lines and orientation of the "plus" 2) How is the "plus" put on the image. Can it be partly off your image? For a first cut approach, assuming you know the exact color red (or even the range of colors), try searching through each pixel in your image until you find a red one. Then you need to see if the shape of your plus is in the appropriate adjoining pixels (this is a series of checks over a specific set of offsets from your "found" red pixel. If you get a match, then you have to check some more to make sure that you are not in a field of red. If your "plus" can be partly off the edge, you will have to compute the remaining portions of the plus and use the approach above with a modified definition of a "plus" This is brute force. There are probably better ways to do this. But this will get you started. Good luck.
-
I know this way but it is a lot of noise in image and the plus isn't like this : ------------o-------- -----------ooo------- ------------o-------- it is like this : -----------------------o----------- -----------------------o----------- -----------------------o----------- ---------------o o o o O --- o o o o ---------------------o------------ ----------------------o----------- ----------------------o----------- ----------------------o----------- and the most important is that it may rotate . and i dont want to rotate it in a ( for ) I want it to be a speedy algorithm . and i also want to use the long tail of plus , I mean it isn't one pixel . thanks -- modified at 10:19 Thursday 11th May, 2006
To find something like that I recommed correlation. When you find your first red pixel, you sum the multiplication of all of the pixels in your defined "plus" with the ones in the appropriate location in the image. Since your image is noisy your defined "plus" needs to be thicker than a single pixel width. Once you get a total sum that exceeds a specified threshold, you declare that a "plus" has been found. Since you have to deal with rotations, the appropriate locations for the multiplies will have to be computed taking different rotations into account. This will increase your processing time significantly. To reduce the number of iterations, once you find the first pixel you know that all other pixels will be right or down from there, assuming the origin is in the top left corner. -- modified at 10:29 Thursday 11th May, 2006
-
I know this way but it is a lot of noise in image and the plus isn't like this : ------------o-------- -----------ooo------- ------------o-------- it is like this : -----------------------o----------- -----------------------o----------- -----------------------o----------- ---------------o o o o O --- o o o o ---------------------o------------ ----------------------o----------- ----------------------o----------- ----------------------o----------- and the most important is that it may rotate . and i dont want to rotate it in a ( for ) I want it to be a speedy algorithm . and i also want to use the long tail of plus , I mean it isn't one pixel . thanks -- modified at 10:19 Thursday 11th May, 2006
You can use 2D-fft (or 2D convolution/correlation). It will work also if there is a rotation on the plus (i.e. there is only a phase rotation) or if there is a scale factor (i.e. the ratio between frequency amplitudes remains constant),..... but they aren't
V_shr wrote:
speedy algorithm
If you want a good and fast way ,probally, your work will be very hard! Try also something using some pre-processing operation, like noise filters and/or colors thresholds(-> RED plus) Hope it helps
-
hello all I need to find a red " + " in a image and i want a good algorithm . thank you -- modified at 9:48 Thursday 11th May, 2006
-
-
hello all I need to find a red " + " in a image and i want a good algorithm . thank you -- modified at 9:48 Thursday 11th May, 2006
Good luck; you'll need it. That's a sophisticated algorithm you're after. I suggest you get a good book on digital processing and don't expect results too quickly. Steve
-
You can use 2D-fft (or 2D convolution/correlation). It will work also if there is a rotation on the plus (i.e. there is only a phase rotation) or if there is a scale factor (i.e. the ratio between frequency amplitudes remains constant),..... but they aren't
V_shr wrote:
speedy algorithm
If you want a good and fast way ,probally, your work will be very hard! Try also something using some pre-processing operation, like noise filters and/or colors thresholds(-> RED plus) Hope it helps
_Russell_ wrote:
Try also something using some pre-processing operation, like noise filters and/or colors thresholds(-> RED plus)
Good!
"Opinions are neither right nor wrong. I cannot change your opinion. I can, however, change what influences your opinion." - David Crow
cheers, Alok Gupta VC Forum Q&A :- I/ IV
-
-
Humm...
"Opinions are neither right nor wrong. I cannot change your opinion. I can, however, change what influences your opinion." - David Crow
cheers, Alok Gupta VC Forum Q&A :- I/ IV
ThatsAlok wrote:
Humm...
pardon dear ? what's up ?
TOXCCT >>> GEII power
[VisualCalc 3.0 updated ][Flags Beginner's Guide new! ]
-
ThatsAlok wrote:
Humm...
pardon dear ? what's up ?
TOXCCT >>> GEII power
[VisualCalc 3.0 updated ][Flags Beginner's Guide new! ]
-
toxcct wrote:
pardon dear ?what's up ?
oops sorry! wrong person :)
"Opinions are neither right nor wrong. I cannot change your opinion. I can, however, change what influences your opinion." - David Crow
cheers, Alok Gupta VC Forum Q&A :- I/ IV