I am working on crops images and want to detect some random pattern from binary image. As in following image want to detect only the area in yellow circle. So how it can be done any algorithm or direct technique? Thanks for you support.

[KALYAN ACHARJYA]

How you got this yellow circles? If you know the center and radius of the circles, you can easily find the area of the circle.

[Image Analyst]

What is special about the blobs in the yellow circles? Why isn't the blob just below the bottom circle one you want to keep? Is it the size, or the brightness, or the "roundness"? What distinguishes those blobs from all the numerous other blobs in the image?

[Asif Ameer]

Really very thanks Image Analyst! in fact this is wheat crop image. and in vertical there are crop rows and encircled portion is weed (unwanted crop) which I want to detect. Here is original image from which I have extracted this image by applying some Excess Green Index & then applying some Binary threshold. Here is original image and conversions. Both actual crop and weed are of same color in case of binary image and want to segment weed only from this output.

[Greg Heath]

I don't know how to do it. However, the obvious difference is the average color level of the surrounding pixels

Hope this helps.

Greg

[Asif Ameer]

well Greg Thanks for your response! But what if I can draw centerline in crop row and then perform some sliding window operation to scan image around this center line and color all these pixels those have been scanned by sliding widow and then perform color based threshold to remove that color area pertaining to crop rows?

Answer 1

[Image Analyst]

I'd use bwareaopen() or bwareafilt() to get blobs that are small. Large blobs will be the big crop rows. Of course when the plants are very young it would be hard to tell between a small wheat plant and a small weed. However John Deere has farm machines that drive over the crop rows and use deep learning to spray only the weeds, so it is a solved problem, and probably one that has patents on the concept.

[Asif Ameer]

Yes definitely most of the image processing problems contains some solution but there is always a a place to solve the problem with different way..... So any of edge detection or sliding window operation that can be used to detect vertical portion of image i.e. (crop rows) further these can be segmented?

Thanks Image Analyst your response is is quite beneficial and helpful.

[Asif Ameer]

or simply creating mask of of crop rows and then segment that all the portion of original binary image by applying mask?

[Image Analyst]

If the rows are not solid vertically, like the plants are still small and growing, then you can get a vertical projection and look for high mean. then threshold to find the rows. Little weeds in between the rows will not be above the threshold.

horizontalProfile = sum(binaryImage, 1);
wheatColumns = horizontalProfile > someValue;

Then you can use regionprops() to find the centroid of the wheatColumns, and again on the original binary image to find the center of all the blobs. Then filter the binary image to accept only those blobs within some x distance of the wheat column centroids. Blobs in between those would be weed blobs and will be not chosen. That's how I'd proceed for cases where the lanes are not solid from top to bottom, like is the case for small, young plants.

I don't see how edge detection would help at all.