Papers by Pradit Mittrapiyanuruk
In this paper, we present an automated system for the visual inspection of cubic zirconia (CZ) cu... more In this paper, we present an automated system for the visual inspection of cubic zirconia (CZ) cut quality. In particular, we inspect the cut quality from pavilion facets of the CZ. For the hardware, the system includes a computerizedcontrol mechanical part that performs both the task of feeding the CZ to the inspection station and the task of separating the gemstone according to the inspection result. In terms of software, we propose an image processing algorithm that consists of two major steps. For the first step, pavilion facets are extracted from the CZ image acquired from the pavilion side. In particular, we resort to the idea of 1-D edge detection in conjunction with random sample consensus line fitting for the pavilion facet extraction. For the second step, a set of measures derived from the extracted facet structure are calculated and are used for cut quality judgment as either accept or reject. The metrological analysis of the system is also investigated. We perform an experiment to inspect 1756 object images consisting of both good and bad samples. The performance of our system yields to about 5.21% of false reject rate and 0% of false acceptance rate. The system can inspect CZ with a rate of 1 sample/s.
⎯In this paper, we present a high speed autofocus system for micro system applications and design... more ⎯In this paper, we present a high speed autofocus system for micro system applications and design a look-up-table based autofocusing algorithm for applications when a target object is always visible, e.g., manufacturing parts with alignment fiducials. We perform an evaluation of 24 focus measures to verify that which focus measure is the best for the look-up-table based method. From the evaluation, we find that the Chebyshev moments-based focus measure (CHEB) is the most suitable. Furthermore, we also develop a look-up-table based autofocus system that uses CHEB as the focus measure. In training phase, we offline construct a table from training images of an object that are captured at several lens distances. Each entry of table consists of focus measure computed from image and lens distance. In working phase, given an input image, the algorithm first computes the focus measure and then finds the best match focus measure from the table and looks up the corresponding lens position for moving it into the in-focus position. Our algorithm can perform autofocusing within only 2 steps of lens moving. The experiment shows that the system can perform high speed autofocusing of micro objects.
We present a people counting algorithm using depth images acquired from a KINECT camera that is i... more We present a people counting algorithm using depth images acquired from a KINECT camera that is installed vertically, i.e., pointing toward the floor. Our proposed algorithm is referred to as Field seeding algorithm. The key idea is that first a set of local minimum values are detected from several spatially distributed seed locations. Then, the people-head blobs are detected from the binary images generated with regard to the threshold values derived from the local minimum values. The recall, accuracy and F-score of our algorithm are comparable to the current state-of-the-art people counting using KINECT, i.e. Water Filling. However, the main advantage over the previous method is that our algorithm operates deterministically, i.e., no any random number generating function is used.
In this paper, we present a single-lens based imaging system that its focal length can be adjuste... more In this paper, we present a single-lens based imaging system that its focal length can be adjusted by using liquid crystal spatial light modulator (LC-SLM). The SLM is placed at the aperture stop of the imaging system to work as a phase mask (PM) that modifies the light wavefront passing the exit pupil. By changing the pixel-based phase mask pattern that sends to the SLM, the focal length of the imaging system can be varied in the same spirit as conventional mechanical-zoom lenses. We demonstrate a proof-of-concept system for high magnification industrial inspection applications in a laboratory setup. We show the results that the focal length of the system is adjustable within the range 2.0mm  .
This paper presents a method for automatic segmentation of blood vessels in retinal images. The m... more This paper presents a method for automatic segmentation of blood vessels in retinal images. The method is based on vessel tracking technique. The key idea of the method is that first a set of seed points (center of vessel cross sections) is extracted. Then, the seed points are connected to establish the vessel skeleton. Finally, the false vessel point are rejected by resorting to a hypothesis-verificaton based procedure. The major contribution of this work is that we formulate the step of seed point connection in the form of graph-theoretical shortest path problem. Then we apply the Dijkstra's algorithm to solve the problem. The performance of our method evaluated on the publicly available DRIVE database shows promising results.
IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, 2013
Proceeding of 4th Symposium …, Jan 1, 2000
Unlike English, there is no explicit sentence marker in the Thai language. Conventionally, space ... more Unlike English, there is no explicit sentence marker in the Thai language. Conventionally, space is placed at the end of sentence in Thai writing. But it does not mean that space always indicates the sentence boundary. It is also used as other purposes ]. This paper presents an algorithm to extract sentences from paragraph by detecting the true sentence breaking spaces, by applying the statistical part-of-speech (POS) tagging technique to the space classification problem. The algorithm considers 2 consequent strings with a space in between each time for determining the space as whether a true sentence breaking space or not. We divided the ORCHID Thai POS tagged corpus into 10 portions for cross-validation test. The evaluation result shows that the average accuracy of space classification and break-space detection are 85.26% and 79.82% respectively and the average of false-break rate is 8.75%. Our approach also shows a significant improvement to the traditional statistical POS tagging technique. The average of POS tagging error rate reduction is as high as 11.3%.
… on Spoken Language …, Jan 1, 2000
This paper presents a method to improve the naturalness of Thai Text-to-speech synthesis, in 4 ma... more This paper presents a method to improve the naturalness of Thai Text-to-speech synthesis, in 4 main parts. In the pausing module, its main function is to determine the break location when synthesizing a Thai text which has no explicit sentence/phrase/word boundary. In the syllable duration and tone generation, a set of rules is provided to generate proper prosodic parameters for synthesizing more natural speech. The syllable duration rule is applied using the Klatt's method to handle the task in syllabic frame. The tonal rule considers the effect of tonal coarticulation and F0 downdrift in generating the F0 contour parameter. In the demisyllable concatenation, the TD-PSOLA technique is applied to modify the waveform for obtaining the required prosody. The LSP-based concatenated boundary smoothing is also included to imitate the crosssyllable coarticulation effect. The result of comparative quality test shows a significant improvement in our proposed method.
Proceedings of the …, Jan 1, 2000
This paper reviews the current state of technology and research progress in the Thai language pro... more This paper reviews the current state of technology and research progress in the Thai language processing. It resumes the characteristics of the Thai language and the approaches to overcome the difficulties in each processing task.
NECTEC Technical …, Jan 1, 2000
Thai -text analysis, prosody generation and speech synthesis. In the text analysis, problems in T... more Thai -text analysis, prosody generation and speech synthesis. In the text analysis, problems in Thai text processing can be decomposed into the models of sentence extraction, phrase boundary determination and grapheme-to-phoneme conversion. The syllable duration and F0 contour generation rules are included in the prosody generation. This is to realize the synthetic speech in the suprasegmental level. In the speech synthesis, the definition and the construction of acoustic inventory structure 'demisyllable' are presented. Furthermore, three signal-processing algorithms, amplitude normalization, the segment boundary smoothing and prosodic modification, are also presented in this topic.
The goal of this communication is to suggest an alternative implementation of the k-way Ncut appr... more The goal of this communication is to suggest an alternative implementation of the k-way Ncut approach for image segmentation. We believe that our implementation alleviates a problem associated with the Ncut algorithm for some types of images: its tendency to partition regions that are nearly uniform with respect to the segmentation parameter. Previous implementations have used the k-means algorithm to cluster the data in the eigenspace of the affinity matrix. In the k-means based implementations, the number of clusters is estimated by minimizing a function that represents the quality of the results produced by each possible value of k. Our proposed approach uses the clustering algorithm of Koontz and Fukunaga in which k is automatically selected as clusters are formed (in a single iteration). We show comparison results obtained with the two different approaches to non-parametric clustering. The Ncut generated oversegmentations are further suppressed by a grouping stage-also Ncut based-in our implementation. The affinity matrix for the grouping stage uses similarity based on the mean values of the segments.
International Journal of Knowledge-based and Intelligent Engineering Systems, 2010
This paper presents a method for tracking the 3D pose of rigid objects. The proposed method is a ... more This paper presents a method for tracking the 3D pose of rigid objects. The proposed method is a 3D extension of the appearance-based approach called Active Appearance Models (AAM). Here, the 3D shape of the object and the geometry of the camera are added as part of the minimizing parameters of the AAM algorithm in order to determine the full 6 degree-of-freedom (DOF) pose of the object. This work is a twofold, major improvement of our previous work: First by applying the inverse compositional algorithm to the image alignment phase; and second, by incorporating the image gradient information into the same image alignment formulation. Both improvements make the method not only more time efficient, but they also increase the tracking accuracy, especially when the object is not rich in texture. Moreover, since our method is appearance-based, it does not require any customized feature extractions, which also translates into a more flexible alternative to situations with cluttered background, complex and irregular features, etc. The proposed method is compared with our previous work and with a previously developed algorithm using a geometric-based approach.
In this paper we present a new method for tracking rigid objects using a modified version of the ... more In this paper we present a new method for tracking rigid objects using a modified version of the Active Appearance Model. Unlike most of the other appearance-based methods in the literature, such as , our method allows for both partial and self occlusion of the objects. We use ground-truth to demonstrate the accuracy of our tracking algorithm. We show that our method can be applied to track moving objects over wide variations in position and orientation of the object -one meter in translation and 140 degrees in rotation -with an accuracy of a few millimeters. . Proceedings of the IEEE Workshop on Motion and Video Computing (WACV/MOTION'05) 0-7695-2271-8/05 $ 20.00 IEEE
This paper presents two different algorithms for object tracking and pose estimation. Both method... more This paper presents two different algorithms for object tracking and pose estimation. Both methods are based on an appearance model technique called Active Appearance Model (AAM). The key idea of the first method is to utilize two instances of the AAM to track landmark points in a stereo pair of images and perform 3D reconstruction of the landmarks followed by 3 D pose estimation. The second method, the AAM matching algorithm is an extension of the original AAM that incorporates the full 6 DOF pose parameters as part of the minimization parameters. This extension allows for the estimation of the 3D pose of any object, without any restriction on its geometry. We compare both algorithms with a previously developed algorithm using a geometric-based approach [14]. The results show that the accuracy in pose estimation of our new appearance-based methods is better than using the geometric-based approach. Moreover, since appearance-based methods do not require customized feature extractions, the new methods present a more flexible alternative, especially in situations where extracting features is not simple due to cluttered background, complex and irregular features, etc.
Computer Vision and Image Understanding, 2004
The goal of this communication is to suggest an alternative implementation of the k-way Ncut appr... more The goal of this communication is to suggest an alternative implementation of the k-way Ncut approach for image segmentation. We believe that our implementation alleviates a problem associated with the Ncut algorithm for some types of images: its tendency to partition regions that are nearly uniform with respect to the segmentation parameter. Previous implementations have used the k-means algorithm to cluster the data in the eigenspace of the affinity matrix. In the k-means based implementations, the number of clusters is estimated by minimizing a function that represents the quality of the results produced by each possible value of k. Our proposed approach uses the clustering algorithm of Koontz and Fukunaga in which k is automatically selected as clusters are formed (in a single iteration). We show comparison results obtained with the two different approaches to non-parametric clustering. The Ncut generated oversegmentations are further suppressed by a grouping stage-also Ncut based-in our implementation. The affinity matrix for the grouping stage uses similarity based on the mean values of the segments.
This paper reviews the current state of technology and research progress in the Thai language pro... more This paper reviews the current state of technology and research progress in the Thai language processing. It resumes the characteristics of the Thai language and the approaches to overcome the difficulties in each processing task.
This paper reviews the current state of technology and research progress in the Thai language pro... more This paper reviews the current state of technology and research progress in the Thai language processing. It resumes the characteristics of the Thai language and the approaches to overcome the difficulties in each processing task.
This paper presents a method to improve the naturalness of Thai Text-to-speech synthesis, in 4 ma... more This paper presents a method to improve the naturalness of Thai Text-to-speech synthesis, in 4 main parts. In the pausing module, its main function is to determine the break location when synthesizing a Thai text which has no explicit sentence/phrase/word boundary. In the syllable duration and tone generation, a set of rules is provided to generate proper prosodic parameters for synthesizing more natural speech. The syllable duration rule is applied using the Klatt's method to handle the task in syllabic frame. The tonal rule considers the effect of tonal coarticulation and F0 downdrift in generating the F0 contour parameter. In the demisyllable concatenation, the TD-PSOLA technique is applied to modify the waveform for obtaining the required prosody. The LSP-based concatenated boundary smoothing is also included to imitate the crosssyllable coarticulation effect. The result of comparative quality test shows a significant improvement in our proposed method.
Uploads
Papers by Pradit Mittrapiyanuruk