This paper proposes a low power tone recognition suitable for automatic tonal speech recognizer (ATSR). The tone recognition estimates fundamental frequency (F0) only from vowels by using a new magnitude difference function (MDF), called vowel-MDF. Accordingly, the number of operations is considerably reduced. In order to apply the tone recognition in portable electronic equipment, the tone recognition is designed using parallel and pipeline architecture. Due to the pipeline and parallel computations, the architecture achieves high throughput and consumes low power. In addition, the architecture is able to reduce the number of input frames depending on vowels, making it more adaptable depending on the maximum number of frames. The proposed architecture is evaluated with words selected from voice activation for GPS systems, phone dialing options, and words having the same phoneme but different tones. In comparison with the autocorrelation method, the experimental results show 35.7% reduction in power consumption and 27.1% improvement of tone recognition accuracy (110 words comprising 187 syllables). In comparison with ATSR without the tone recognition, the speech recognition accuracy indicates 25.0% improvement of ATSR with tone recogntion (2,250 training data and 45 testing words).

This paper presents a technique for the variation of hysteresis band in delta-sigma modulation. A sinusoidal, and a random hystersis band are combined to achieve an optimal performance in terms of constant switching frequency and the harmonic spikes. The sinusoidal hysteresis band technique produces a constant switching frequency while the random hysteresis band suppresses the harmonic spikes. The effects of various variations of hysteresis band on the harmonic spectrum characteristic were described. The technique is experimented in a single-phase inverter and the harmonic peaks and the distortion of output voltage were used to measure the performance of the proposed technique.

In order to navigate a mobile robot or an autonomous vehicle in indoor environment, which includes several kinds of obstacles such as walls, furniture, and humans, the distance between the mobile robot and the obstacles have to be determined. These obstacles can be considered as walls with complicated edges. This paper proposes a mobile-robot-navigation method by using the polar coordinate transformation from an omnidirectional image. The omnidirectional image is obtained from a hyperboloidal mirror, which has the prominent feature in sensing the surrounding image at the same time. When the wall image from the camera is transformed by the transformation, the straight lines between the wall and the floor appear in the curve line after transformation. The peak point represents the distance and the direction between the robot and the wall. In addition, the wall types can be classified by the pattern and number of peak points. They are one side wall, corridor and corner. To navigate the mobile robot, in this paper, it starts with comparing a peak point obtained from the real image with the reference point determined by designed distance and direction. If there is a difference between the two points, the system will compute appropriate wheel angle to adjust the distance and direction against the wall by keeping the peak point in the same position as the reference point. The experiments are performed on the prototype mobile robot. The results show that for the determining distance from the robot to the wall between 70-290 cm, the average error is 6.23 percent. For three types of the wall classification, this method can correctly classify 86.67 percent of 15 image samples. In the robot movement alongside the wall, the system approximately consumes the 3 frame/s processing time at 10 cm/s motion speed. The mobile robot can maintain its motion alongside the wall with the average error 12 cm from reference distance.

Bone Mineral Density (BMD) is an indicator of osteoporosis that is an increasingly serious disease, particularly for the elderly. To calculate BMD, we need to measure the volume of the femur in a noninvasive way. In this paper, we propose a noninvasive bone volume measurement method using x-ray attenuation on radiography and medical knowledge. The absolute thickness at one reference pixel and the relative thickness at all pixels of the bone in the x-ray image are used to calculate the volume and the BMD. First, the absolute bone thickness of one particular pixel is estimated by the known geometric shape of a specific bone part as medical knowledge. The relative bone thicknesses of all pixels are then calculated by x-ray attenuation of each pixel. Finally, given the absolute bone thickness of the reference pixel, the absolute bone thickness of all pixels is mapped. To evaluate the performance of the proposed method, experiments on 300 subjects were performed. We found that the method provides good estimations of real BMD values of femur bone. Estimates shows a high linear correlation of 0.96 between the volume Bone Mineral Density (vBMD) of CT-SCAN and computed vBMD (all P<0.001). The BMD results reveal 3.23% difference in volume from the BMD of CT-SCAN.

Automatic detection of normal mammograms, as a "first look" for breast cancer, is a new approach to computer-aided diagnosis. This approach may be limited, however, by two main causes. The first problem is the presence of poorly separable "crossed-distributions" in which the correct classification depends upon the value of each feature. The second problem is overlap of the feature distributions that are extracted from digitized mammograms of normal and abnormal patients. Here we introduce a new Support Vector Machine (SVM) based method utilizing with the proposed uncrossing mapping and Local Probability Difference (LPD). Crossed-distribution feature pairs are identified and mapped into a new features that can be separated by a zero-hyperplane of the new axis. The probability density functions of the features of normal and abnormal mammograms are then sampled and the local probability difference functions are estimated to enhance the features. From 1,000 ground-truth-known mammograms, 250 normal and 250 abnormal cases, including spiculated lesions, circumscribed masses or microcalcifications, are used for training a support vector machine. The classification results tested with another 250 normal and 250 abnormal sets show improved testing performances with 90% sensitivity and 89% specificity.

In retinal blood vessel extraction through background removal, the vessels in a fundus image which appear in a higher illumination variance area are often missing after the background is removed. This is because the intensity values of the vessel and the background are nearly the same. Thus, the estimated background should be robust to changes of the illumination intensity. This paper proposes retinal blood vessel extraction using background estimation. The estimated background is calculated by using a weight surface fitting method with a high degree polynomial. Bright pixels are defined as unwanted data and are set as zero in a weight matrix. To fit a retinal surface with a higher degree polynomial, fundus images are reduced in size by different scaling parameters in order to reduce the processing time and complexity in calculation. The estimated background is then removed from the original image. The candidate vessel pixels are extracted from the image by using the local threshold values. To identify the true vessel region, the candidate vessel pixels are dilated from the candidate. After that, the active contour without edge method is applied. The experimental results show that the efficiency of the proposed method is higher than the conventional low-pass filter and the conventional surface fitting method. Moreover, rescaling an image down using the scaling parameter at 0.25 before background estimation provides as good a result as a non-rescaled image does. The correlation value between the non-rescaled image and the rescaled image is 0.99. The results of the proposed method in the sensitivity, the specificity, the accuracy, the area under the receiver operating characteristic (ROC) curve (AUC) and the processing time per image are 0.7994, 0.9717, 0.9543, 0.9676 and 1.8320 seconds for the DRIVE database respectively.

In tooth contour extraction there is insufficient intensity difference in x-ray images between the tooth and dental bone. This difference must be enhanced in order to improve the accuracy of tooth segmentation. This paper proposes a method to improve the intensity between the tooth and dental bone. This method consists of an estimation of tooth orientation (intensity projection, smoothing filter, and peak detection) and PCA-Stacked Gabor with ellipse Gabor banks. Tooth orientation estimation is performed to determine the angle of a single oriented tooth. PCA-Stacked Gabor with ellipse Gabor banks is then used, in particular to enhance the border between the tooth and dental bone. Finally, active contour extraction is performed in order to determine tooth contour. In the experiment, in comparison with the conventional active contour without edge (ACWE) method, the average mean square error (MSE) values of extracted tooth contour points are reduced from 26.93% and 16.02% to 19.07% and 13.42% for tooth x-ray type I and type H images, respectively.

This paper proposes a new controlling technique of asynchronous sigma delta modulation with characteristic of one-cycle response. This technique can reject power source perturbations in one cycle period and reduce the peaks of harmonic with one side random hysteresis technique. The proposed method was analyzed, designed, and experimented in a full bridge inverter. The distortion of output voltage and the harmonic peaks were used to measure the performance of the proposed technique. The experimental results show that the proposed technique can reduce the peak of harmonic up to 0.42 p.u and the harmonic distortion 5.9% at the ripple of 20% of power source when comparing with convention asynchronous sigma delta modulation.

In developing an automatic system of a single tooth length measurement on x-ray image, since a tooth shape is assumed to be straight and curve, an algorithm which can accurately deal with straight and curve is required. This paper proposes an automatic algorithm for measuring the length of single straight and curve teeth. In the algorithm consisting of control point determination, curve fitting, and length measurement, PCA is employed to find the first and second principle axes as vertical and horizontal ones of the tooth, and two terminal points of vertical axis and the junction of those axes are determined as three first-order control points. Signature is then used to find a peak representing tooth root apex as the forth control point. Bezier curve, Euclidean distance, and perspective transform are finally applied with determined four control points in curve fitting and tooth length measurement. In the experiment, comparing with the conventional PCA-based method, the average mean square error (MSE) of the line points plotted by the expert is reduced from 7.548 pixels to 4.714 pixels for tooth image type-I, whereas the average MSE value is reduced from 7.713 pixels and 7.877 pixels to 4.809 pixels and 5.253 pixels for left side and right side of tooth image type-H, respectively.

This paper presents a method for detection of calcification, which is an important early sign of breast cancer in mammograms. Since information of calcifications is located in inhomogeneous background and noises, it is hard to be detected. This method uses wavelet packet transform (WPT) for elimination of the background image related to low frequency components. However, very high frequency signals of noises exist with the calcifications and make it hard to suppress them. Since calcification location can be represented as vertical, horizontal, and diagonal edges in time-frequency domain, the edges in spatial domain can be utilized as a filter for noise suppression. Then the image from inverse transform will contain only required information. A free-response operating characteristic (FROC) curve is used to evaluate a performance of proposed method by applying it to thirty images of calcifications. The results show 82.19 percent true positive detection rate at the cost of 6.73 false positive per image.