The book is divided into two parts. The first part is about advances in machine learning, pattern recognition, data clustering and graph matching.

Nine interesting theoretical contributions are presented in the first section. Two articles cover the problem of graph matching; one focuses on attribute relational graphs and presents a combination of two existing search-algorithms (exhaustive and genetic) tested on artificially created data. The second one discusses the comparison and recognition of objects, and it presents a different vector extraction based distance function called energy function, but to me, the results are not that clear.

Machine learning is the common topic in the other articles. The first article of this section presents profound theoretical results on the problem of estimating probabilities using feedback for the binomial and multinomial case. Other ones focus on the possible practicability of such methodology in different scenarios. Training of supervised pattern recognition is used for detecting malicious intrusions in network traffic through raw packets from tcpdump traces. Training of supervised neural networks is applied for the prediction of membrane protein structure using existing protein datasets and binary encoding schemas of amino acid sequences. An ensemble learning based model is introduced for the problem of new expert addition and old expert retirement in pattern recognition under concept drift, with diversity measures based criteria for decision making. Comparison of learning algorithms for feature extraction of infrasound signals using neural networks deals with discrete wavelet transforms, time scale spectra, and cepstral coefficients and their derivatives.

Another two articles in the first section cover related subjects. Symmetry-based clustering is applied using a new distance measure for indicating the appropriateness of datasets in a validity index. Prediction engineering and a risk limitation model for quantifying investment risk in the stock market are tested using historic data and suggestions for implementation are given.

The second part of the book has sixteen articles about advances in biometrics with pattern recognition and data mining. All of them present interesting applications, extensions and modifications to existing methodologies.

Linguistics and character recognition research are found in six documents:

1. Lexicon-based algorithm labeling anomalous documents for detecting potentially criminal behavior (terrorist activities) from data in web documents,

2. Artificial neural network for Ethiopic character recognition trained with string patterns to handle character variations,

3. One-stroke character recognition using a directional features recognition method,

4. Support vector machine for number plate recognition,

5. Two semi-supervised learning methods and one statistical hidden Markov model are evaluated for the named entity recognition system used for Bengali language identification,

Offline segmenting hand-written Farsi/Arabic overlapped or connected words for automatic text recognition, using a large database of pre-processed handwritten Arabic words.

Six documents cover the recognition in biometrics via video and audio. An extensive paper presents a generic learning machine in convolutional neural networks for face image processing, used in face detection, facial feature detection, face alignment, gender classification, and face recognition. Other papers deal with pattern recognition and clustering of facial thermal features for classifying affecting states; optimization of principal component analysis by reducing the dimension of images for face recognition; clustering of skin pixels for training a face detection and recognition classifier in order to discriminate unknown faces and using a probability vector based filter; simplification of iris identification algorithm for the implementation in low cost devices, without compromising its recognition capabilities; and a learning machine in the form of a coupled hidden duration semi Markov model for conversational audio data analysis and its classification.

Finally, scenery and image analysis is covered in four papers.

1) Pattern recognition for 2D barcode PDF417 reading and processing using a CCD camera and not the conventional laser scanning devices.

2) Binarization for image processing of cheques in Persian language using Otsu and Background Subtraction algorithms, and trained with a database of 150 cheque images.

3) Audio and video fusion for indoor and outdoor scene recognition with the purpose of its classification, and a learning machine trained with a database of sampled videos taken from a digital video camera.

4) A decision tree method for identification of horror movies based on shot-length and motion intensity features obtained from video analysis, for the intelligent indexing of multimedia database.

Although in my opinion many articles could have presented more proper conclusions or deeper proofs and evidences, and some of them focused on the practicability of machine learning and pattern recognition from a theoretically point of view, the scientific relevance of the content of the book is good. The authors presented their work at the International Workshop on Advances in Pattern Recognition 2007. Accordingly, the target audience is also academic. .

The book would have benefitted from correcting some editing errors and grammatical mistakes, though.

Click here for Top of Page
Right Arrow: Next
Right Arrow: Previous



Progress in Pattern Recognition


by Sameer Singh and Maneesha Singh (Eds.)

Springer, 2007

Series:  Advances in Pattern Recognition


Reviewed by

Eleazar Jimenez Serrano (Japan)

Click on the image (above) to go to the publisher’s web page for this book where you will find a description of the book.