Tool to help Count Features in Images

Craic has released a new tool to help count features in images. This can be used in a range of applications such as counting buildings in satellite images, bacterial colonies and cell types in histology images.


It takes the form of a Ruby Sinatra application that fetches remote images and Javascript in the client web page that implements the marking and counting functions.

The user can choose the shape and color of the marker. The current total is updated as features are clicked and the coordinates of each point are stored internally.

The image with the user's marks can then be saved to a PNG format file and the list of coordinate pairs can be displayed in a separate window.

The graphics and interaction are implemented with Canvas and Javascript and make use of the wonderful Canvas2Image code from Jacob Seidelin which allows the current state of a canvas element to be saved to an image file.

The live application is hosted at Heroku and can be accessed at http://counter.craic.com

The code is distributed freely under the terms of the MIT license and is archived at Github

Counting Temporary Shelters in Satellite Images using OpenCV

A number of areas of data analysis are dominated by sophisticated algorithms, intensive computation and, in some cases, limited access to raw data. Examples include the analysis of satellite imagery, feature extraction from video, protein structure analysis and language processing.

I am interested in how simple, approximate methods can be used to extend the application of these technologies. While simple approaches will clearly not match the accuracy and resolution of complex methods, they can, by nature of their simplicity, be implemented and deployed more easily and hence more widely.

I have just posted the code for one of these projects to the Craic Github site. The application involves image processing of satellite images downloaded from Google Maps in order to estimate the number of temporary dwellings or shacks in the slums of Port-au-Prince in Haiti.

The satellite images of Port-au-Prince show large areas covered with small white, blue and rust colored squares. These are the slums of the city in areas such as St.Martin, Cite Soleil and others. 

The blue features are very distinctive and most likely represent the ubiquitous blue plastic tarpaulins the you can find at any hardware store. 

Using the Python interface to the wonderful OpenCV image processing library, I wrote a simple application that identifies blue features below a cutoff size and then computes their area. By fetching adjacent squares that tile across the city, and calculating the area of blue features, I can come up with an matrix showing the relative density of the slums.

The approach is undoubtedly simplistic (the code is only about 35 lines of Python) but it demonstrates how simple approaches can be successfully applied to complex and sophisticated types of data. You do not always need access to expensive commercial software and proprietary datasets in order to work in these fields.

The code is made freely available at https://github.com/craic/count_shelters and you can read my write up of the work so far, with example images HERE.