A. Baskar

Automatic Book Spine Extraction and Recognition for Library Inventory Management

Manual inventory management in a library is by far arduous. Automation of book inspection can be achieved by using a simple camera based system that can recognize book spines in a book shelf. The book spines contain printed information such as title, author and publisher name, which can be extracted and verified with the librarys database. Book spines can be segmented by detecting their rectangular boundaries which appear as straight lines. Line detection using hough transform and line segment detector may result in spurious boundaries due to the presence of long titles or graphics on the book spine. In this paper, we propose a technique to improve book spine border detection by devising set of constraints based on structural properties that can be used to filter the detected line segments so as to obtain book spine boundaries. The segmented book spines are binarized to extract the printed information such as title, author and publisher name. The text is recognized using Tesseract Optical Character Recognition Engine. The proposed algorithm was tested successfully on book shelf images with vertically oriented, uniformly inclined and multi-oriented book spines.

Improved Adaptive Skip Algorithm for Video Shot Boundary Detection

Video shot boundary (VSB) detection plays an important role in modern video archiving and summarization applications. The classical linear approach towards VSB detection is basically a sequential search, however this approach is computationally very expensive for large databases. In this work we propose a non-linear approach for VSB detection, which is an improvement of existing adaptive skip. The proposed technique compares less than 10% of frames in an entire video. Our experimental result shows that the proposed technique outperforms the other existing non-linear algorithms with better accuracy and meaningful VSBs

An intelligent system of content-based image retrieval for crime investigation

Crimes in the modern world have been increasing dramatically in proportion with technological growth. In many situations, investigators have a sketch of the criminal, generated from the details given by the eye witnesses. There is a need of a better methodology for identifying the culprits from those images. Nowadays, RAM and memory sizes have become huge and in turn, there is a tremendous increase in the size of the databases. Investigators have huge datasets with details, photographs, images of the criminals or crime scenes and this is the second important consideration. Image search has become highly challenging task in these dense database environments. Accuracy and speed are two important factors for this type of image searches. We propose a compact embedded search engine to search and extract images from these data bases, using a content-based image retrieval CBIR algorithm.

Facial Expression Classification Using Machine Learning Approach: A Review

Automatic Facial Expression analysis has enthralled increasing attention in the research community in excess of two decades and its expedient in many application like, face animation, customer satisfaction studies, human-computer interaction and video conferencing. The precisely classifying different emotion is an essential problem in facial expression recognition research. There are several machine learning algorithms applied to facial expression recognition expedition. In this paper, we surveyed three different machine learning algorithms such as Bayesian Network, Hidden Markov Model and Support Vector machine and we attempt to answer following questions: How classification algorithm used its characteristics for emotion recognition? How various parameters in learning algorithm is devoted for better classification? What are the robust features used for training? Finally, we examined how advances in machine learning technique used for facial expression recognition?