neural networks is the best approach....since all ships differ only in minute details.The correlation between features will be high.SVM is the best starting point in terms of computation speed and simplicity but accuracy depends a lot on the images training set.on the other hand neural networks are computationally slow( the parallel programming using CUDA or so could help you for real time approach). There is not really a specific algorithm that works well, the choice of algorithms depends on training set and on many factors....

Neural networks? Is this real time? I dont think that it is real time.

You can try doing some PCA on a feature vector and from there chose the features that best recognize your ships.

Which features?

I think, length, wides, effect of ships on water is better.

Which features do u have suggest?

I think you need to try standart computer vision technic of recognition object. Good start is HOG features and SVM (nonlinear, because ships have many views). Or BOW of SURF and SVM. And of case you need big data set of ships and detector of ships. (The simple way is use bacground substraction).

yes Mr. Sergieskiy. substraction is good to find ship in images, but how to recognize of type of ships? which feature must be extracted?

Identifying anti-warships is easy: look for a greenpeace flag on it!

You should, first, understand which are the features that allow to disciminate the differences between ships (for instance how you, seeing the images can discriminate the kind of ship?). After this you can start in finding how to describe, geometrically or mathematicaly, such differences. Only after this human-guided parameter extraction you can use machine vision or AI-based techniques.

Erfan Amani

I Suggest use learning features of images and dont use handmade features.

First you need to ctreate learnin database:

1. Get many videos with ships.

2. Get annotation (position and type)

2.1. If you (your team) are able to annotating ships (use label me video http://code.google.com/p/labelme-video/)

2.2. Else you can search for open annotating ships (http://www.image-net.org/)

2.3. Create own annotating tool

2.3.1 Download video with one type of ships.

2.3.2 Use background substraction

2.3.3. Cut rectangle images from frame

2.3.3. Filter that images

2.4. Resize all images to one size (with proportion: fix width=const)

3. Extract Feature

3.1. HOG

3.2. SURF

3.3. Patch

3.4. just image

4. Learn classifier

4.1. SVM

4.2. boosting

4.3. random forest

4.4. neural network

5. Profit

What is the resolution of the different ships with respect to the whole image (pixel count)?. Is the satellite geostationary, that you can use a baseline landmark for relative sizing?