clasifiy kitchenware using neural networks: What are the features

hi

i try to use neural network to build simple system that classify four element

Screw, Spoon, Knife, Thorn

my problem is in feature extraction

This is the first application to me So my experience a few in feature extraction

can any help in chose good feature

[Walter Roberson]

I have not heard of a "thorn" in this context. Do you have a link to an example image?

[Image Analyst]

Another name for chopstick perhaps??? And screw = corkscrew?

http://www.mathworks.com/matlabcentral/answers/7924-where-can-i-upload-images-and-files-for-use-on-matlab-answers

[mangood UK]

dear Walter Roberson

i will take this as example

five class

[mangood UK]

or these

both i can work on but

i need some help to start

Answer 1

[Image Analyst]

Features could be any of the things that regionprops() returns. Most likely each of the features (e.g. area, perimeter, solidity, majoraxis length, etc.) is in a certain range for each type of kitchen implement, as long as you don't have lots of different sizes (like a dozen different sizes and styles of knives). See my tutorial for how to extract features from objects in an image: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

[mangood UK]

dear image analyst

is their an algorithm to find length of specific object??

like in binary image the object is white and background black

how to find length or maximum width of such object???

[Walter Roberson]

It is not possible to find the length of an object without having some calibration information available; without that, the most you can do is find the distance in pixels. Any object can be made to occupy the same number of pixels distance as any other object, by changing the magnification.

[Image Analyst]

See my image segmentation tutorial: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862.

[mangood UK]

dear: Walter Roberson ..thank you

-----------------------------

dear: Image Analyst

thank you for your tutorial its amazing

-------------------------------------------------

but i have idea and i need your opinion

for screw objects, second pic above

i will use number of pixel that the object take as one of the feature

the second feature is to take edge detection of object then take mean of it

i use this second feature because i note that from pic that edge difference between objects!!!!

best regards

[Image Analyst]

OK, whatever works. Just be aware that your items are shiny and thus the number of edges in the image depends on what is being reflected by the surfaces, so a screw or spoon might have 9000 "edge" pixels at one location and then if you just move it to a new location it may have 3000 or 15000 edge pixels because it's reflecting different things around it. Things like area and perimeter length are more robust to things like illumination and clutter surrounding the objects.

[mangood UK]

waw .....

thank you for share your experience.

i read your tutorial and now i can find the area in pixel but for perimeter is there a function in matlab that do this function

for last do you have a book ???

and please if you have time and not busy

make to us tutorial for object feature extraction

like previous tutorial with advanced method

best regards:

m.s

[Image Analyst]

You can ask for 'Perimeter' in regionprops().

I don't have a book but Steve does: http://imageprocessingplace.com/DIPUM-2E/dipum2e_main_page.htm. My tutorial will work for more features than in my image segmentation tutorial. There are other things that regionprops() can measure that I didn't ask for. Look in the help. Of course you can also invent your own features. For example regionprops doesn't output the very useful metric "circularity" for some reason, but I can calculate it from the equation

circularities = perimeters.^2 / (4*pi*Areas).

I was just discussing with a development team from the Mathworks yesterday that I should package up my demos and upload them. I have well over 100 demos that I've developed into general purpose demos in response to questions here in Answers - I've just never found the right time to zip them all up and upload them, not to mention that I add or change files on an almost daily basis. Maybe someday...

[mangood UK]

the demo that by you is much useful than book so i ,and other as i think wait to put your package as example in matlab example which will make matlab very rich information for academic use and other use

 best regards
m.s