Actually, shapes alone are not enough to recognize the content of the image with high accuracy, ideally, you need to use another information such colors and textures etc, however, you can see the following links for more information:

http://muvis.cs.tut.fi/Documents/MscThesisPartio.pdf

https://arxiv.org/ftp/arxiv/papers/1005/1005.5437.pdf

http://www.engineersgarage.com/contribution/content-based-image-retrieval-matlab-project

 edge detection and linking algorithms  may help you. Gradient vector flow and morphological operations can also be used to detect shape from an image.  Also , Moments are used to extract shape from an image. hu and zernike moments are few example.

This a simple demonstration of a content based image retrieval using 2 techniques.

1. Using knn for image retrieval

2. Using svm for image retrieval.

For further details check these links :

http://www.mathworks.com/matlabcentral/fileexchange/42008-content-based-image-retrieval

http://www.engineersgarage.com/contribution/content-based-image-retrieval-matlab-project