Warning: The display option can only plot 2D training data

I am working on classification of images using SVM classifier,

I want to do classifivation ,the dataset consists of males and females ,first i want to classify its gender,

if it is female i dont want to classify,if male i want to calssify it as adult or aged person

for this i have 100 images of males and females

i have extracted features like colour and texture and have saved in variable features

features =100x6,(100 images ,each column consists of feature vector),

features1 i have to test for 40 images(40x6)

so on whic basics i have to classify using svm classifier

please assist

tr=rand(100,6);features

group=[ones(60,1);2*ones(40,1)];

te=tr(61:end,:);

svmStruct = svmtrain(xdata,group,'showplot',true); res= svmclassify(svmStruct ,te);

I get error in plotting,plz assist.

Answer 1

[Shashank Prasanna]

You have high dimensional data (in |R^6) and obviously you can't visualize 6 dimensions, so don't force the 'showplot' to true and you shouldn't see this 'Warning'

If you are concerned about visualization take the dimensionality reduction route and perform for example PCA and reduce it to 3 dimensions to visualize it.

[FIR]

CAn you tell how to perform it

[FIR]

after performing svmclassify and performing tabulate(),i get 40% 1's,60% 2's,why i get only 6o% accuracy

[Walter Roberson]

It is not possible to classify gender from photographs.

[FIR]

walter i have extracted features ,even then it is not possible to perform?

[Walter Roberson]

Classifying gender by photographs is about like trying to classify religion by photographs.

[FIR]

OK Waltet can i relace it by medical image dataset,classifying normal or abnormal,if abnormal then classifying again

[Walter Roberson]

You could try.

What kind of images would they be, and what would "abnormal" signify?

If you are trying to classify cells as normal or cancerous by shape, then typical result accuracy rates seldom exceed 78%, usually lower (64% to 72%). You need more complex classification algorithms to reach 78% by visuals, and are quite unlikely to exceed that for visual-light images. To get higher you need to switch out of visible-light measurements, such as switching to infrared, or Magnetic Resonance.

[FIR]

Walter i have taken mri imag of brain,how can i classify it as normal or cancer,can i extract glcm features

[FIR]

and regionprops for shape,like area,centroid etc

[Walter Roberson]

We gave up on classifying mri brain images as cancerous or not, well over a decade ago. It is a multi-year project. You pretty much need to have access to the Brain Atlas and do non-linear image registration in order to figure out which part of the brain you are looking at and then to match the locations of the brain to known functional areas. Once you have that, then you can start comparing the actual images of the area to the typical images so that you can figure out whether particular areas are supposed to be high or low intensity, with a view to finding something that should not be there. And figuring out whether it is cancer or just something "odd" (e.g., you want to ignore metal nails embedded in the brain because metal nails are unusual but not cancer.) All this with an mri image that might only be 64 x 64.

A guided search is much easier: that is, to compare the images to the Brain Atlas still, and find the areas that are statistically different and highlight those areas for interpretation by the technician or doctor. Instead of making a decision about them, just say "This part needs a closer look" because I don't know what is going on here."

We moved away from MRI imaging for cancer classification, to go to MRS, Magnetic Resonance Spectroscopy, which is a chemical analysis. You can do a lot more with chemical analysis!

[Walter Roberson]

? I thought FIR was working on this, not nkumar ?

[FIR]

walter i have wxtracted glcm features for MRS image and have stores in variable D,on which basics can i classify it as normal abnormal,i have to use svmclassifier

[Walter Roberson]

MRS does not produce images; it produces floating point spectra, which glcm would not be suitable for.

MRI is used for images, and as I indicated before there are a number of difficult steps in figuring out where in the brain you are and what should be in that area, which is needed before you can use features to decide whether you have anything abnormal.

[FIR]

ok walter if i have extrated feature values say area this is just for an example,using svm how i can classify it is normal or abnormal.

[Walter Roberson]

You cannot do that without information about what should be in that area. In some places in the brain, near blackness is what you would normally expect, and anything substantially different would be "abnormal", but in other places in the brain, darkness is what would be "abnormal"

Suppose I show you a picture of a rock in my back yard. The rock is wet.

• Question: Is that abnormal?
• Answer: if the picture was taken in the middle of summer, dry would be typical
• But July is the month we get the most rain (statistically), so a wet rock in Summer is not what you would usually see, but is not unusual
• But if the picture was taken in the middle of our winter, then a wet rock outside would be abnormal, as our winters are quite cold and the water would have frozen within a couple of minutes rather than stayed liquid.

You need context to interpret results, context of what is usual in that area. Without that context, your program might as well be full of rocks.

[FIR]

thanks a lot for your suggestions walter,is there any demo for any images how classification is made,I want to know the concept of how classification is made