The goal of our project is to deliver an automatic recognition system of Venetian Palaces from the pictures. We are making a web application which takes an image of arbitrary Venetian building and it returns the name of the corresponding building. In our previous blog, we described about the Venetian architecture and their characteristics details. We also described about the pipeline of our project.
In the web application part we have learned how to use Django web development framework. We have created database model for storing and retrieving image files for processing using OpenCV. In Django, we have created an admin interface, where we as the admins can add the images, their names, and other relevant information to our database. We have only added a few images to the database. These images in the database can be retrieved, opened and modified using the functions of OpenCV in python. Some basic functions of converting the images from colour to black/white was also tested. In this way we have tested the integration of OpenCV libraries on our web application platform. We are now confident that this web application platform can handle image processing.
We are on the process of creating the user side of our website. For this we need to use HTML and CSS files embedded with python objects as on a typical Django application. The next milestone in this part of the project would be to complete the web application for the user side and also improve the kind of information that we want to present the user upon completion of the image recognition. We will add and show the location of the palace on a map, how to reach there; among other relevant details.
Palace Recognition: Phase II
For making a recognition model, we require a large collection of good quality images of the Venetian buildings. In the past 4 weeks, we successfully collected more pictures of Venetian buildings of around 60 palaces. After cropping the palaces from the scanned book images, we moved onto the second phase of the project of pre-processing and feature extraction from these images. We face a problem that though we have large number of categories but we managed to collect less than 10 images for each category. So, we tried to increase the dataset collection by dividing each image into random patches of images. Precisely, each image is made into 6-7 images. All the images were converted to 256×256 images to make it uniform and applied whitening. Now, we are presently trying to extract different features such as SIFT, HOG Descriptors, etc to find the closest match to architectural feature. We have the problem of choosing a particular learning model. We are working with different models like Random forest, SVM, pattern matching, etc. We will try to resolve the learning model in the next phase.
In this phase, we increased the collection of dataset and finished the pre-processing of images in Python. Now, we are working on feature extraction and learning model simultaneously, in MATLAB. We plan to write the code in Python after getting the suitable learning model for our problem. In the next phase, we shall also focus on developing the user side of the website and integrating image processing with it.
. Rössler, Jan-Christoph. I Palazzi Veneziani. Venezia: Fondazione Giorgio Cini, 2010. Print.
. Fasolo, Andrea. Palazzi Di Venezia. Venezia: Arsenale, 2003. Print.
. Venetian Architecture. (n.d.). Retrieved March 4, 2015, from http://imaginingvenice.com/2013/04/16/venetian-architecture/