Dong Sun Park

Leukocyte image segmentation by visual attention and extreme learning machine

This paper presents a fast and simple framework for leukocyte image segmentation by learning with extreme learning machine (ELM) and sampling via simulating visual system. In sampling stage, visual attention and the effect of microsaccades in fixation are simulated. The high gradient pixels in fixation regions are sampled to group training set. We designed an automatic sampling process for leukocyte image according to the staining knowledge of blood smears. In learning stage, ELM classifier is trained online to simulate visual neuron system and then extracts pixels of object from image. The ELM-based segmentation is fully automatic by the proposed framework, which could find efficient samples actively, train the classification model in real time and almost no parameter adjusted. Experimental results demonstrated the new method could extract entire leukocyte from complex scenes, has equivalent performance compared to the SVM-based method and exceeds the marker-controlled watershed algorithm.

Applying learning vector quantization neural network for fingerprint matching

A novel method for fingerprint matching using Learning Vector Quantization (LVQ) Neural Network (NN) is proposed. A fingerprint image is preprocessed to remove the background and to enhance the image by eliminating the LL4 sub-band component of a hierarchical Discrete Wavelet Transform (DWT). Seven invariant moment features, called as a fingerCode, are extracted from only a certain region of interest (ROI) of the enhanced fingerprint. Then an LVQ NN is trained with the feature vectors for matching. Experimental results show the proposed method has better performance with faster speed and higher accuracy comparing to the Gabor feature-based fingerCode method.

An Optimal Orientation Certainty Level Approach for Fingerprint Quality Estimation

Analyzing the quality of fingerprints in advance can be benefit for a fingerprint recognition system to improve its performance. Representative features for the quality assessment of fingerprint images from two existed types of capture devices are different. Orientation certainty level (OCL) is an effective method to extract image orientation feature. However it is not an effective estimation system to cooperate with the extracted features. In this paper, we explore the application of optimization theory, and support vector machine (SVM) in the field of image processing. Our proposed optimal orientation certainty level (OOCL) approach calculates the OCL for each block, extracts features from the optimal OCL system and uses the SVM classifier to determine whether an image should be accepted as an input to the recognition system. Experimental results show that the proposed OOCL method can improve the recognition rate than OCL method.

Fingerprint Matching Using Invariant Moment Features

A method for fingerprint matching using invariant moment features is proposed. The fingerprint image is first preprocessed to enhance the original image by the Short Time Fourier Transform (STFT) analysis. Then, a set of seven invariant moment features is extracted to represent the fingerprint image from a Region of Interest (ROI) based on the reference point of the enhanced fingerprint image. The reference point is determined by the complex filters method. Finally, a Back Propagation Neural Network (BPNN) is trained with the features for matching. Experimental results show the proposed method has better performance with higher accuracy and faster speed comparing to the traditional Gabor feature-based fingerCode method.

Fingerprint reference point detemination based on a novel ridgeline feature

Different with a traditional orientation analysis or frequency analysis which is commonly used for the fingerprint reference point determination, a new fingerprint reference point determination method based on a novel inconsistency feature obtained from the relationship between ridgelines and curves of a fingerprint is proposed. For the proposed method, an improved enhancement method is introduced and the posterior probability theory is used to determine the reference point in pixel precision. Experimental results demonstrate its feasibility, validity and ability independent of the quality and types of fingerprints.

Rule-Based Fingerprint Quality Estimation System Using the Optimal Orientation Certainty Level Approach

The fingerprint quality can be used as a good predictor for fingerprint recognition performance. Knowing the fingerprint quality in advance is useful to improve the performance of fingerprint recognition system. In this paper, we propose an effective quality estimation system with 4 rules which are applied in two-step. Each rule consists of a Back-Propagation Neural Networks (BPNN) classifier based on Optimized Orientation Certainty Level (OOCL) features extracted locally from fingerprint images. Experimental results show that the proposed two-step OOCL method can estimate fingerprint quality more effectively.

A blind watermarking method based on cellular automata transform domain

A blind watermarking method using cellular automata transform (CAT) is proposed. A binary watermark is embedded in a certain sub-band, which is further subdivided into non-overlapped blocks, of 2-level CAT transformed of a host image. Each block embeds a watermark bit with two patterns (P0, P1) only differing in sign. The detector is blind, determining the value of a watermark bit on the sign of the correlation between the corresponding block and one of the two patterns. Three keys are used to enhance the security of the proposed scheme. The proposed method has a reasonable degree of watermark imperceptibility, which can be notified from the high PSNR and MSSIM. The results show that the proposed method is robust against most common image operations.

Fingerprint Reference Point Determination Based on Orientation Features

In this paper, a new orientation-based method which operates in two-step, called in Dynamic Processing (DP) system, is proposed for the determination of fingerprint reference point. For the two-step operation, it uses different orientation features at different scales: block-based orientation certainty and pixel-based segmented direction map. A block-based orientation certainty is used to describe the change of a block curvature of a fingerprint, which is determined by two eigenvalues of the gradient covariance matrix and a pixel-based segmented direction map is used to find intersections emerging from the directional transition. The DP system is built through their cross-references to determine the position of reference point. While the proposed system has the pixel-based precision by virtue of using a pixel-based segmented direction map, it reduces many possible fault symptoms by virtue of using a block-based orientation certainty. The proposed technique shows better performance in accuracy rate than other previous techniques. The performance of the proposed one is verified through simulations and its analysis.

Microcalcifications detection in digital mammogram using morphological bandpass filters

Microcalcifications in digital mammogram images can be an early sign of breast cancer. It is very challenging, however, to detect all microcalcifications since they appear as slightly brighter spots than their backgrounds. In this paper, a new method is proposed to efficiently detect all microcalcifications using morphological bandpass filters. Morphological bandpass filters, each with two different structuring elements, are tuned to isolate frequency components of microcalcifications in this method. Experimental results show that the proposed method with bandpass filters can recognize microcalcifications with a higher visibility and more accurate positions and sizes comparing to the well-known wavelet transform method.

New mechanisms for end-to-end security using IPSec in NAT-based private networks

While the transition from IPv4 to IPv6 has been considered to extend the IP address space, the NAT protocol is widely used as an interim solution. Using the NAT protocol with the end-to-end IPSec resulting a conflict due to the address transition operation of the NAT. In this paper, we design two mechanisms which provide the end-to-end security service even if a NAT is used for private networks. The first proposed mechanism defines a notification message to deliver the address translation information in advance. This mechanism uses already defined protocols and does not need additional protocol modification. The second proposed mechanism uses SSL and IPSec to protect user data and IP header. Although this mechanism needs chip redundancy on packet length, it can save duplicated encryptions caused by SSL and IPSec encryptions. Procedures and parameters to support the mechanisms are designed in this paper.