Ask Your Question
0

Is there any programs to recognize objects?

asked 2013-03-13 01:55:57 -0500

hor gravatar image

HI

IS there any programs that can recognizing objects? I mean after a car was in front of the camera the program says car or after a bird did it , it says bird .

Also I want to know what is the name of this field of image processing or i can search what WORDs or PHRASEs for find more information or instruction for it?

Thanks a lot

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted
2

answered 2013-03-13 03:26:53 -0500

Considering you have to find an object and an image, there is already a large difference in approaches:

  1. Object recognition: look at an image and classify the complete image as a single label. With recognition, the object that needs to be detected is known beforehand, like the exact shape.
  2. Object detection: you actually want to detect an object class inside input images. In brief you need to build a model for this object and run this model over the complete images.

Both techniques have different approaches, but some that might be usefull according to the exact problem you are facing are:

  1. bag of words approach
  2. k - nearest neighbours approach
  3. object detection using a cascade of weak classifiers
  4. support vector machine object detection
  5. random forest object detection

Some of these techniques depend on a large set of labeled images, assigning a class, then defining the nearest example to the new test image and feeding it the same label. Other techniques use tons of positive and negative object sampels to create a abstract model of how the item should look like.

When looking for identical known objects or models, template matching could be used.

Some techniques require the detection of features in the image, which could be extracted using specific image processing techniques. Examples are HAAR wavelets, Local Binary Patterns, Histogram Of Oriented Gradients, ...

Matching does require you to find interesting points, and describe them in a discrete way, this can be done by using feature descriptors.

So basically, go through the following information and describe an exact situation you want to reach:

Also some guides to basic recognition and detection can be found here:

Hope this helps you determine the right solution for your problem!

edit flag offensive delete link more

Comments

Thanks a lot dear friend I didn't see the links since now but I will read them soon . My purpose is that I want to describe some objects for program (like car, tree , street lights ,etc) and when the camera see those (in every directions) the program can recognize them .

I think I understand whole of solution but I must read more about details and I think building the template models should be my approach , but i am not sure!

And also i couldn't understand your mean in this paragraph :

Some of these techniques depend on a large set of labeled images, assigning a class, then defining the nearest example to the new test image and feeding it the same label. Other techniques use tons of positive and negative object sampels to create a abstract model of how the item should look like.

hor gravatar imagehor ( 2013-03-13 04:18:05 -0500 )edit

Can you explain above paragraph? Thank you in advance !

hor gravatar imagehor ( 2013-03-13 04:19:16 -0500 )edit

What i mean there are basically two approaches. First one with the nearest exampel label is sometimes used for face recognition, an example based approach. It means that you take tons of images of people, like 50 of each person, create an abstract representation of them (for example eigenfaces or fisherfaces) and store them into a database. Then for a new input, create same abstract representation and use a distance measure towards the database. The most resembling example in the database is used to identify the new sample. The second approach i mention is more general. It takes for example 5000 faces and tries to find general characteristics. From those characteristics it defines a model of how a face should look like. For example when using HOG, gradients are used to define a face.

StevenPuttemans gravatar imageStevenPuttemans ( 2013-03-13 04:28:43 -0500 )edit

At a new entry, the same characteristics are retrieved, looking if it matches the model. Then a detection is performed. Does this makes it more clear? If not, maybe read through the Viola&Jones paper, describing the cascade of weak classifiers approach for face detection. Link : http://research.microsoft.com/en-us/um/people/viola/Pubs/Detect/violaJones_CVPR2001.pdf

StevenPuttemans gravatar imageStevenPuttemans ( 2013-03-13 04:30:22 -0500 )edit

Do not forget that the thing you try to do is one of the most hottest topics in todays computer vision conferences. I am doing some kind of the same research in that field, you can always check my project website http://www.eavise.be/tobcat/. But I am implementing those algorithms based on OpenCV now :)

StevenPuttemans gravatar imageStevenPuttemans ( 2013-03-13 04:31:54 -0500 )edit

Thanks a lot dear friend

I found a youtube movie it's exactly that I want .pls look at it and tell your opinion about what approach is used there and is there that program code's? or something like that?

http://www.youtube.com/watch?v=uwNO2Hl6cuU

hor gravatar imagehor ( 2013-03-16 06:27:05 -0500 )edit

@StevenPuttemans thank you for the explanation, I have a small question to ask: from a video capture and a set of images (sqlite) I want to set the image that is closest to it I detect matches the two images and then using a technique such as Histogram to define closest distance

Marwen Z gravatar imageMarwen Z ( 2013-03-31 05:05:22 -0500 )edit
StevenPuttemans gravatar imageStevenPuttemans ( 2013-03-31 09:46:08 -0500 )edit

And @hor, what you are looking for is feature based matching. Go to the opencv tutorials and look there for some information, follow the tutorials to learn what they doing in the video: http://docs.opencv.org/doc/tutorials/features2d/table_of_content_features2d/table_of_content_features2d.html#table-of-content-feature2d

StevenPuttemans gravatar imageStevenPuttemans ( 2013-03-31 09:48:29 -0500 )edit
Login/Signup to Answer

Question Tools

Stats

Asked: 2013-03-13 01:55:57 -0500

Seen: 3,785 times

Last updated: Mar 13 '13