Hello all, I am new to image processing.... I want to detect edges from an image. When I searched on detecting edges in images I came across about edge detection using genetic algorithms. Where I find the chromosomes and find fitness function for crossover and mutation process. Also, they say that chromosomes are binary array. I am finding it difficult to understand how can we find binary arrays from an image and picking that as a chromosomes or do we have any methods to find chromosomes from an image. Please help me...any help is appreciated.
Hello, I have no experience in genetic algorithms but may be able to help anyway: from your description I've get the idea about what to do.
1. From your introduction I assume you have not yet tried edge detection the "classic" way. I'd suggest to get a feeling for the task itself by first implementing classic edge detection (with the usual filters).
2. Now, what about genetic algorithms ? Start with a 2D array (I would not recommend to use a binary array - try the whole thing in 8 bit gray scale) which is initialized "somehow".
Apply it on sample images.
Until you get "chromosomes" (aka 2 D arrays) with at least marginal success apply random changes to the elements of the chromosome(s).When you have gotten chromosomes with some success, crossover would be implemented as "overlaying" chromosomes - weighted adding of the array elements of 2 chromosomes.
The tricky part is the fitness function: I'd suggest to use the results from classic edge detection on the sample images and compare them with the results from the application of a chromosome. Fitness would be expressed in how good the genetic result is - compared to the classic detection.
Quite an interesting field of application. And a lot of work (and literature research) to do. Enjoy - and have fun!
Dear Dinesh,
Thank you for posting. We have a great deal of experience in this area and have tried to make these problems easy to solve with our new software MIPAR, available at http:://MIPAR.us. We don't implement genetic algorithms, but do have an optimization engine where process parameters may be objectively determined using image similarity metrics such as mutual information. I'd be very happy to recommend algorithms, or better yet, put together a MIPAR recipe file for you and send it over, if you could just post an example image. Looking forward to hearing from you.
Cheers,
John
Thank you for your time Dreher. your first point is right. In the second point you said that "start with 2-D array initialized "somehow" and apply it on sample images"... can you please elaborate it bit more on "initializing somehow" and "applying". what do you mean by applying on sample images?...i may sound silly but please help me.
Hello,
"somehow" could be any pattern of your like. Or some random initialization. The genetic evolution theory does not make any assumptions about the start element.
"applying" is about testing your chromosome by executing the actions necessary to detect the edges in your sample images. In the classic algorithms is is called "folding" the "filter" (aka array) with the image. More graphic, it is like sliding your array over the image to subsequently cover all input pixels. If you do not get it: imagine inspecting an image by means of a magnifying lens that is smaller than your image.
@Dreher so do you mean to say i'll take this matrix as my initial chromosome [1,2,3;4,5,6] and slide it across all over the image is it right?
Thank you so much for your time Dreher.. Now i'll try to implement and let you know if i have any doubts.
1.Taking gray levels of histogram of an image in an array u can create a chromosome.
2. Taking coordinate (i, j) and amplitude of pixels in an array u can do it.
Generate an expression including statistical parameters and features of image as the fitness function.
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Enzymology
- Enzymes