Detecting "brown spots" on leafs

Can you show me will I be able to detect the "brown spots" in this image? Thanks

Answer 1

[Image Analyst]

I'd first define a mean green color. Then convert to lab color space and calculate a delta E image. Then threshold the delta E to find out where the delta E is more than some certain amount. Those will be the non-green pixels. See my delta E demo at http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862 and adapt it. I do this kind of thing all the time and yours is not that hard. Note, you cannot use automatic thresholding despite what some novices at image processing may suggest because you need to find all amounts of brown from 0% to 100%.

[Elvin]

Actually, I'm trying to detect three types of plant disease, these are the images.

Bacterial Leaf Blight

Brown Spot

Rice Blast

Should I use the same method (the one you posted above) for the three diseases? Can you show me the methods of detecting each disease.

Thank you very much.

[Image Analyst]

Yes. The delta E will detect all color differences. Once you have identified the spots that are not green, you can do analysis of them to determine which disease they are based on their color, pattern, or whatever is it that distinguishes one disease from another.

[Elvin]

I'm confused again. I have the ff questions:

1. I'd first define a mean green color. Then convert to lab color space and calculate a delta E image

- What do you mean by "define a mean green color"? Also, how will I be able to get the delta E of the image if I only have one LAB channels?

2. Note, you cannot use automatic thresholding...

- Can you show me how to do it manually? I never used thresholding before.

3. Once you have identified the spots that are not green, you can do analysis of them to determine which disease they are based on their color, pattern, or whatever is it that distinguishes one disease from another.

- Can you suggest an analysis for identifying the disease?

[Image Analyst]

Um, kind of looks like you're wanting me to write the whole app for you. I'll give you some pointers - that's all I have time for.

1. Convert to lab color space. Then you need to define a "healthy" or "normal" green. Try looking at a leaf with no disease and finding out what the mean LAB is. Then calculate the delta L, deltaA and deltaB for every pixel in the image to that reference green. Square them add them and take square root to get the delta E. You will get a monochrome image where high values are disease and low values are normal healthy tissue.

2. Threshold at some value, like 5 or something

diseasedPixels = deltaEImage > 5; % a binary image of diseased areas.

3. I don't know what the different diseases are. Maybe some give white spots and some give red spots. Or some give round spots and some give slivers. I don't know but I think you must know what is different. All diseases don't look the same do they? If not, then identify what is different and measure that.

[Elvin]

1. First I need to mask the leaf part only for more accuracy right? I have this code in my nitrogen before:

LeafImage = imread('test1.jpg');
leaf_mask = LeafImage(:,:,2) > 32;

will that mask only the green part of the leaf or it will also include the brown/white spot present on the leaf?

2. thanks for that. I'll try it.

3. Do you think shape detection is enough for this?

[Image Analyst]

1. That could work. Try it and see. 3. I have no idea. Could be color, could be texture, could be shape, could be size. You're the one with the different disease images, not me, so I have no idea. I would think you would know. What do you see that is different between the different diseases?

[Elvin]

1. The code for masking works only for green. The leaf contains also different color such as brown and white. How should I mask the leaf containing all the colors present in the leaf?

Can I just compare the histograms for identifying the disease?

[Image Analyst]

It should work for any color channel, as long as that color channel's pixels are brighter than 32. If you get holes, then call imfill(binaryImage, 'holes'); or bwareaopen().

[Elvin]

I tried the imfill but there are still holes in my mask.

By the way, what do you mean by * You will get a monochrome image where high values are disease and low values are normal healthy tissue*? When I tried using imshow(deltaE), I've got nothing. I've tried the method of getting the deltaE of a healthy leaf and a diseased leaf. What's next with that?

[Walter Roberson]

Try

imshow(deltaE,[])

[Elvin]

Still nothing :(

Here's my code:

clc
clear

HealthyImage = imread('1.jpg');
cform = makecform('srgb2lab');
HealthyImage = applycform(HealthyImage,cform);
LChannel_HealthyImage = mean2(HealthyImage(:, :, 1));
AChannel_HealthyImage = mean2(HealthyImage(:, :, 2)); 
BChannel_HealthyImage = mean2(HealthyImage(:, :, 3));

RBImage = imread('RB - 1.jpg');
cform = makecform('srgb2lab');
RBImage = applycform(RBImage,cform);
LChannel_RBImage = mean2(RBImage(:, :, 1));
AChannel_RBImage = mean2(RBImage(:, :, 2)); 
BChannel_RBImage = mean2(RBImage(:, :, 3));

deltaL = LChannel_HealthyImage - LChannel_RBImage;
deltaA = AChannel_HealthyImage - AChannel_RBImage;
deltaB = BChannel_HealthyImage - BChannel_RBImage;

deltaE = sqrt(deltaL^2 + deltaA^2 + deltaB^2);
imshow(deltaE,[])

[Image Analyst]

deltaE is a 2D array that has a color difference value it it, just like any other 2D numerical array. Look at deltaE in the variable inspector to see the values or use imshow like Walter showed. If you don't know what the variable inspector is, then see this.

[Image Analyst]

You need to do an element by element squaring:

deltaE = sqrt(deltaL.^2 + deltaA.^2 + deltaB.^2);

Answer 2

[sai rama krishna gondela]

can any one write complete program for brown spots or tthe above program is itself correct

[Image Analyst]

Probably. And I think you're one of those people.

Answer 3

[Kameshwar Pramanik]

can you tell me how to correct following line i am getting error on this. That is File "RB - 1.jpg" does not exist.

for Code is: RBImage = imread('RB - 1.jpg');

http://in.mathworks.com/matlabcentral/answers/110329-detecting-brown-spots-on-leafs#comment_186504

[Image Analyst]

Make sure the filename is spelled correctly and the file is in the current folder or on the search path.

[swathi]

Hello.. Even I am looking for similar sort of code, on the mango images. finding spot pixels. I tried the same code, but its not wworking, Could somebody please help.