Computationally assessing the visual qualilty of a web page

• Main Author: Janardhanan Nambiar, Aparna
• Other Authors: School of Electrical and Electronic Engineering
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/53052
• Institution: Nanyang Technological University

Though there are millions of websites on the internet, half of the ones we come across do not provide an enjoyable user experience to people. This could be in the sense of usability, functionality, aesthetics or credibility. The aim of this thesis is to apply and improve algorithms for image recognition on web pages to assess their visual quality. Techniques such as Spatial Pyramid Matching, Support Vector Machines, etc. have been successfully applied for feature matching and classification. In this thesis, the author tries to apply this approach to web pages under the assumption that a similar pattern or a feature set could be extracted from the image of a web page. In other words, the aim is to obtain an image classifier with the feature set which can be used to assess web pages. For this a database of web page images was created and labelled as aesthetically appealing and non-appealing based on ground truth from a number of subjects. These were then classified into training and testing datasets. Features were first extracted from the training set and used for dictionary creation, spatial pyramid matching and SVM training. Feature extraction and pyramid matching were done on the testing data as well. Following this a number of experiments were run to find the best values for parameters such as weightage for various features, cost, pyramid levels and dictionary size. The analysis of these results produced some expected and some surprising results that could have an impact on the use of image recognition techniques for assessment of web page quality.