Luan Ling Lee

A prototype for Brazilian bankcheck recognition

This paper describes a prototype for Brazilian bankcheck recognition. The description is divided into three topics: bankcheck information extraction, digit amount recognition and signature verification. In bankcheck information extraction, our algorithms provide signature and digit amount images free of background patterns and bankcheck printed information. In digit amount recognition, we dealt with the digit amount segmentation and implementation of a complete numeral character recognition system involving image processing, feature extraction and neural classification. In signature verification, we designed and implemented a static signature verification system suitable for banking and commercial applications. Our signature verification algorithm is capable of detecting both simple, random and skilled forgeries. The proposed automatic bankcheck recognition prototype was intensively tested by real bankcheck data as well as simulated data providing the following performance results: for skilled forgeries, 4.7% equal error rate; for random forgeries, zero Type I error and 7.3% Type II error; for bankcheck numerals, 92.7% correct recognition rate.

Disconnected handwritten numeral image recognition

Describes a method for numeral character recognition. Initially, the image of an unknown numeral is pre-processed, and two feature sets are compiled and used for numeral character recognition. The first feature set is compounded by topological characteristics and by characteristics obtained from a pictorial distribution analysis of numeral images. The second feature set is the proper set of numeral images, after being normalized. The classification process is divided into two stages. In the first stage, the classification is based on the first feature set. In the second stage, Hopfield networks are used to find the most probable numeral class. Experimental results obtained from testing laboratory-prepared data and handwritten numerals extracted from real Brazilian bank checks show that recognition rates of 85% and 92.4% were achieved, respectively.

An off-line method for human signature verification

This paper presents an off-line human signature verification system. The following information was used to characterize a static signature for verification purpose: the slant directions of the signature strokes and those of the envelopes of the dilated signature images. The system employing a linear classifier provides an equal error rate of 14.6% for skilled forgeries, and 3.0% for simple forgeries.

Automatic extraction of filled information from bankchecks

Presents a technique for extracting the filled information from bankchecks. We have analyzed the bankchecks characteristics and propose a model that can be used to locate and extract the filled information of any bankcheck. This model is based on prior knowledge about check layout structure and in check identification through reading the magnetic ink character recognition line. The redundant information, such as the background pattern, and printed lines and characters, is eliminated from the check image. We maintain a sample of the checks background pattern stored and use it in order to eliminate the original background pattern from digitized checks. Next, the areas where the filled information is supposed to appear are extracted through a template. The baselines are eliminated using an approach based on projection profiles, while the printed characters are eliminated through a subtraction operation. Experimental results from testing real Brazilian bankchecks show that the proposed method is capable of extracting the filled items from bankchecks as an accuracy rate of 88.7-98.3%.

Compression of bank cheque images based on layout knowledge

In this paper a scheme for bank cheque images compression based on layout knowledge is proposed. The layout structure of the cheques is analyzed and the nonessential parts are located. These parts, viz., the background and the printed information, are eliminated from the original image. The resulting image contains some noise that are eliminated by a filtering operation. The image is enclosed to eliminate some uninformative parts. The final image has only the filled information. The digitized image can be easily reconstructed by restoring the filled information and summing it with background and printed information. The proposed compression scheme is tested by Brazilian bank cheques. Comparisons with other compression schemes, shows that the proposed scheme performs significantly better in terms of the compression efficiency, maintaining the visual quality.

Automatic storage, retrieval and visualization of bank check images

This paper presents an automated system for storage and retrieval of bank checks in contrast with the microfilming techniques that are currently used. The bank check images are introduced into an extraction module where the filled in information is segmented. This information is indexed via keywords derived from the MICR line and stored in a database under a hybrid structure where hash tables, trees and inverted files are employed. For the information retrieval and visualization, make-up bank check images are generated. The experimental results reveal a good performance of the proposed method in terms of compactness of stored information and high visual quality of the reconstructed images.

Principal Component Analysis for Symmetric Key Generation

This work presents a novel biometric encryption scheme based on feature vectors extracted from a face recognition system. This system uses principal component analysis, in order to generate a symmetric secret key, being this key used to encrypt any information data, like a biometric template. The data is therefore concealed and only an individual having a similar biometric feature vector is capable to regenerate the correct key. This scheme is applied to a system using eigenfaces for recognition, where the corrected detected class from a sample image can guarantee the corrected generation of a symmetric key. Due to the efficiency of the system being dependent of the face recognition algorithm, the tests showed a rate of 90.4% of corrected symmetric key generation, or sucessfull encryption/ decryption scheme, for 25 face classes, with 5 images each.

Automatic Extraction of Filled-in Information from Bankchecks Based on Prior Knowledge about Layout Structure

This paper presents a technique for extracting the filled-in information from bankchecks based on prior knowledge about their layout structure. We have analyzed the bankcheck characteristics and proposed a model that can be used to locate and extract the filled-in information applicable to any bankcheck. The model is based on prior knowledge about the check layout structure and on the identification of the check by reading the information stored in the MICR line. To eliminate the redundant information from a bankcheck image, such as the background pattern, the printed lines and the printed characters, we perform as follows. First of all we subtract the digitized check image from the checks background pattern image which is previously stored in the recognition system. Then the areas where the filled-in information is supposed to appear are extracted through a template. The elimination of the baselines in the image is based on projection profiles, while the printed characters are eliminated through a subtraction operation. Experimental results from testing Brazilian bankchecks show that the proposed method is capable of extracting the filled-in items from bankchecks achieving accuracy rates varying from 88.7% to 98.3%.

Typing Biometrics User Authentication based on Fuzzy Logic

This paper uses a fuzzy logic approach in a static typing biometrics user authentication. The inputs are the down and up times, and the ASCII code of the keys that are captured while the user is typing a known string. In this research, it was collected four features (the key code, two keystroke latencies and the key duration) captured in two different strings. The first string was imposed, and the second one was chosen by each user. Seven experiments were developed utilizing a fuzzy logic classifier and the proposed features. The results of the experiments are evaluated in three situations of authentication: the legitimate user, the impostor and the specialist impostor. The best results were achieved utilizing all the features, obtaining a false rejection rate of 3.5% and a false acceptance rate of 2.9%. This approach can be used in the usual login-password authentication for improvement of the false acceptance rate, when the password is no more a secret.In this paper recurrent neural networks are considered to realize traffic prediction in computer network.

Automatic classification of deformed handwritten numeral characters

Describes a method which utilizes Hopfield neural nets to classify those handwritten numerals presenting deformations and stylistic traces. Information for the classification consists of some topological image features and the image pixel distribution. If the recognition cannot be done by these features due to noise and deformations in the images of the numerals, the classification process is performed by four Hopfield neural nets. Using four such nets, we are able to minimize the problem caused by correlated patterns, and also to increase the neural classifiers pattern storage capacity. The proposed method was tested on 121 Brazilian bank checks, achieving a 92.4% correct recognition rate.