Ask Your Question

How to detect(rotation scale invariant) a insect from a picture?

asked 2013-01-09 05:06:19 -0600

this post is marked as community wiki

This post is a wiki. Anyone with karma >50 is welcome to improve it.

Help! I am trying to detect certain insect from a picture or a video.As is well known,Haar,LBP and HOG are tested well for faces or pedestrian detection.But the insect is often rotated in reality while faces and pedestrian are not.So I urgently need a suitable method,such as certain feature or algorithm, to finish my work.If you have dealt with this similar problem, please give me some inspiration.I will appreciate you so much.

Here are some examples: image description image description

edit retag flag offensive close merge delete


Please post some example images. Insects may be really small, texture-less and significantly rotated which would pose a problems to some feature detectors.

Adi gravatar imageAdi ( 2013-01-10 12:40:34 -0600 )edit

Thanks for your attention,Adi. Some examples are added. My main task is to find the objects in a picture. I have used features of Haar, HOG and LBP to train samples. During the training period, the detection rate is at least 0.98 and the falsealarm rate has decreased to 0. But the result is not good. There are too many false-counted regions. So I feel these features may be unsuitable for this work. Do you have any advice to solve the problem? Please give me some suggestion. Thank you!

Kobe隐身 gravatar imageKobe隐身 ( 2013-01-10 23:37:43 -0600 )edit

1 answer

Sort by » oldest newest most voted

answered 2013-01-12 12:38:26 -0600

this post is marked as community wiki

This post is a wiki. Anyone with karma >50 is welcome to improve it.

One solution For this problem ,you can use rotation & sclae invariant template matching for example SURF, FREAK or BRISK descriptors and flann matcher by LSH param.

See there how to use it.

Another solution for this problem is saliency detection & zernik moment .the saliancey detection is used for localization of object then use this position by the snake active contour for find contour of object then you zernik moment for classify(reconstruct) of contour.

edit flag offensive delete link more


Thanks for your suggestion. My main task is to find the objects in a picture, just like face detection, not recognition. I am a rookie in this field. I am getting up to speed and trying to read as much as I can. According to some literature, the methods you mentioned are usually used for edge detection or recognition. I do not confirm these approaches can be used to detect objects. Moreover, because just started getting in touch with object detection, I have never used these arithmetic operators before. Could you tell me how to realize my idea in detail?

Kobe隐身 gravatar imageKobe隐身 ( 2013-01-13 03:05:45 -0600 )edit

Question Tools

1 follower


Asked: 2013-01-09 05:06:19 -0600

Seen: 2,639 times

Last updated: Jan 12 '13