Content-based indexing of low resolution documents

Md Nor, Danial (2016) Content-based indexing of low resolution documents. Doctoral thesis, Universiti Tun Hussein Onn Malaysia.


Download (1MB) | Preview
[img] Text (Copyright Declaration)
Restricted to Repository staff only

Download (269kB) | Request a copy
[img] Text (Full Text)
Restricted to Registered users only

Download (5MB) | Request a copy


In any multimedia presentation, the trend for attendees taking pictures of slides that interest them during the presentation using capturing devices is gaining popularity. To enhance the image usefulness, the images captured could be linked to image or video database. The database can be used for the purpose of file archiving, teaching and learning, research and knowledge management, which concern image search. However, the above-mentioned devices include cameras or mobiles phones have low resolution resulted from poor lighting and noise. Content-Based Image Retrieval (CBIR) is considered among the most interesting and promising fields as far as image search is concerned. Image search is related with finding images that are similar for the known query image found in a given image database. This thesis concerns with the methods used for the purpose of identifying documents that are captured using image capturing devices. In addition, the thesis also concerns with a technique that can be used to retrieve images from an indexed image database. Both concerns above apply digital image processing technique. To build an indexed structure for fast and high quality content-based retrieval of an image, some existing representative signatures and the key indexes used have been revised. The retrieval performance is very much relying on how the indexing is done. The retrieval approaches that are currently in existence including making use of shape, colour and texture features. Putting into consideration these features relative to individual databases, the majority of retrievals approaches have poor results on low resolution documents, consuming a lot of time and in the some cases, for the given query image, irrelevant images are obtained. The proposed identification and indexing method in the thesis uses a Visual Signature (VS). VS consists of the captures slides textual layout’s graphical information, shape’s moment and spatial distribution of colour. This approach, which is signature-based are considered for fast and efficient matching to fulfil the needs of real-time applications. The approach also has the capability to overcome the problem low resolution document such as noisy image, the environment’s varying lighting conditions and complex backgrounds. We present hierarchy indexing techniques, whose foundation are tree and clustering. K-means clustering are used for visual features like colour since their spatial distribution give a good image’s global information. Tree indexing for extracted layout and shape features are structured hierarchically and Euclidean distance is used to get similarity image for CBIR. The assessment of the proposed indexing scheme is conducted based on recall and precision, a standard CBIR retrieval performance evaluation. We develop CBIR system and conduct various retrieval experiments with the fundamental aim of comparing the accuracy during image retrieval. A new algorithm that can be used with integrated visual signatures, especially in late fusion query was introduced. The algorithm has the capability of reducing any shortcoming associated with normalisation in initial fusion technique. Slides from conferences, lectures and meetings presentation are used for comparing the proposed technique’s performances with that of the existing approaches with the help of real data. This finding of the thesis presents exciting possibilities as the CBIR systems is able to produce high quality result even for a query, which uses low resolution documents. In the future, the utilization of multimodal signatures, relevance feedback and artificial intelligence technique are recommended to be used in CBIR system to further enhance the performance.

Item Type: Thesis (Doctoral)
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7800-8360 Electronics > TK8300-8360 Photoelectronic devices (General)
Divisions: Faculty of Electrical and Electronic Engineering > Department of Electronic Enngineering
Depositing User: Miss Afiqah Faiqah Mohd Hafiz
Date Deposited: 01 Sep 2021 07:56
Last Modified: 01 Sep 2021 07:56

Actions (login required)

View Item View Item