Automatic Recognition of Trademark Logos using Computer Vision and Image Processing Techniques

Robust stereo in video sequences using semi-locally recognition. White dots corresponds to the matched points: the moving figure in the foreground is avoided

Nowadays one of the issues software development factories are trying to achieve is a system that could retrieve similar images given a particular one. There are a big family of situations in where a tool that could reduce the number of images for a user to inspect will be saving greats amounts of company time. One of this cases is find similar trademark logos. This is a serious matter for big companies who has to figth the plagiarism.

We present here a tool that allows you to make differents kinds of searchs among a database of images. The tool provides a simple way to retrieve images based on their shapes characteristics. It creates a SQL database with specific image features for content based image retrieval (CBIR). Furthermore simple methods for searching the database and result browsing are provided.

Software Provided

The main product is a C++ library(Windows/Linux compatible) with a friendly interface which provides a series of useful functionalities :

  • Image recognition with semi-local features

    A deep searching technique allows us to find out if some parts of a logo are inside another one.

  • Image recognition with global features

    For fastest results another searching technique is provide. It provides a quick review of all the images that are globally similar to the query.
Both provides the user functions to add a new image and different functions for search similar images in the current database. Both features have a set of parameters that allow the tunning of the system for a particular set of images, in this case trademark logos.

The system has been tested against a database of over 8000 images of real-world trademark logos and against a academic wellknown database(MPEG7 CE 2 Part B).

Green background : query ;Ren background : last really look a like image; Blue background : similar image

Example of image recognition with global features