Tsuyoshi Kawaguchi

Iris detection using intensity and edge information

Abstract In this paper we propose a new algorithm to detect the irises of both eyes from a face image. The algorithm first detects the face region in the image and then extracts intensity valleys from the face region. Next, the algorithm extracts iris candidates from the valleys using the feature template of Lin and Wu (IEEE Trans. Image Process. 8 (6) (1999) 834) and the separability filter of Fukui and Yamaguchi (Trans. IEICE Japan J80-D-II (8) (1997) 2170). Finally, using the costs for pairs of iris candidates proposed in this paper, the algorithm selects a pair of iris candidates corresponding to the irises. The costs are computed by using Hough transform, separability filter and template matching. As the results of the experiments, the iris detection rate of the proposed algorithm was 95.3% for 150 face images of 15 persons without spectacles in the database of University of Bern and 96.8% for 63 images of 21 persons without spectacles in the AR database.

Detection of eyes from human faces by Hough transform and separability filter

We propose a new algorithm to detect the irises of both eyes from a human face in an intensity image. Using the separability filter proposed by Fukui et al., the algorithm first extracts intensity valleys, which we call blobs in this paper, as the candidates for the irises. Next, for each pair of blobs, the algorithm computes a cost using the Hough transform and separability filter to measure the fit of the pair of blobs to the image. Then, the algorithm selects a pair of blobs with the smallest cost as the irises of both eyes. As the result of the experiment using all faces without spectacles in the face database of the University of Bern, the success rate of the proposed algorithm was 96.5% on the average.

Minimizing Total Completion Time in a Two-Machine

We consider the problem of minimizing total completion time in a two-machine owshop. We present a heuristic with worst-case bound 2β/(α + β), where α and β denote the minimum and maximum processing time of all operations. Furthermore, we analyze four special cases: equal processing times on the first machine, equal processing times on the second machine, processing a job on the first machine takes time no less than its processing on the second machine, and processing a job on the first machine takes time no more than its processing on the second machine. We prove that the rst special case is NP-hard in the strong sense and present an O(n log n) approximation algorithm for it with worst-case bound 4/3. We repeat the easy polynomial algorithms for the cases two and three, and show that problem four is solvable in polynomial time as well.

Automatic eye detection using intensity and edge information

We propose a new algorithm to detect the pupils of both eyes from a human face in an intensity image. First, feature points which are the candidates for the pupils of both eyes are extracted from the face image by using the feature template proposed by Lin and Wu (see IEEE Trans. Image Processing, vol.8, no.6, p.834-45 (1999). Next, the proposed algorithm computes a cost for each pair of feature points satisfying a spatial constraint. The cost is computed by searching for a circular region corresponding to the iris around each feature point. Finally, the algorithm determines a pair of feature points with the smallest cost to be the pupils of both eyes. As a result of the experiment using all faces without spectacles in the face database of the University of Bern, the success rate of the proposed algorithm was 93.0% on the average. Also, if looking-down faces are excluded, the success rate of the proposed algorithm was 97.1% on the average.

Recognition of occluded objects by a genetic algorithm

In this paper we present a new algorithm to extract and locate partially visible objects from a 2D image. For each model object, the algorithm cuts a prominent fragment front its boundary. Next, using a genetic algorithm, the algorithm finds an image fragment with the best match to the model fragment from match boundary of the image objects. As a result, the corresponding model is extracted and located from the image. The proposed algorithm can be used for the recognition of objects whose scales, rotations and translations are unknown. In addition, the algorithm can locate objects whose boundaries are distorted.

Minimizing Total Completion Time in a Two-Machine Flowshop: Analysis of Special Cases

We consider the problem of minimizing total completion time in a two-machine flowshop. We present a heuristic with worst-case bound 2β/(α +β), where α and β denote the minimum and maximum processing time of all operations. Furthermore, we analyze four special cases: equal processing times on the first machine, equal processing times on the second machine, processing a job on the first machine takes time no more than its processing on the second machine, and processing a job on the first machine takes time no less than its processing on the second machine. We prove that the first special case is NP-hard in the strong sense and present an O(n log n) approximation algorithm for it with worst-case bound 4/3; we show that the other three cases are solvable in polynomial time.

Eye detection based on grayscale morphology

In this paper we propose a new algorithm to detect eyes in intensity images. The algorithm can detect eyes directly from the whole image. In the algorithm, eyes and mouth are modeled by upright ellipses and cheeks are modeled by circles. The algorithm first detects a valley map of the intensity image using grayscale morphology. In the valley map, eyes and mouth have many valley pixels while cheeks have few valley pixels. In addition, the movement of eyes and mouth decreases valley pixels inside them while the movement of cheeks increases valley pixels inside them. Using these properties of facial features, the algorithm detects two circles corresponding to cheeks and three ellipses corresponding to eyes and mouth. As the results of the experiments using 252 faces without spectacles in the AR face database show, the eye detection rate of the proposed algorithm was 98.8%.

Robust extraction of eyes from face

In this paper we propose a new algorithm to detect the irises of both eyes from a human face in an intensity image. Using the separability filter proposed by Fukui et al. (1997), the algorithm first extracts intensity valleys, which we call blobs in this paper, as the candidates for the irises. Next, for each pair of blobs, the algorithm computes a cost using Hough transform and separability filter to measure the fit of the pair of blobs to the image. And then, the algorithm selects a pair of blobs with the smallest cost as the irises of both eyes. As the result of the experiment using all faces without spectacles in the face database of the University of Bern, the success rate of the proposed algorithm was 96.5% on average.

Detection of target models in 2D images by line-based matching and a genetic algorithm

In this paper we propose an algorithm to detect a 2D target model in cluttered environments and to find its scale s, translation (tx, ty) and orientation θ. The algorithm first extracts straight line segments from the image. Next, generating transformation vectors (s, tx, ty, θ), the algorithm computes the fitness values for these vectors through matching between data line segments extracted from the image and model line segments transformed by (s, tx, ty, θ). And, by a generic algorithm, the algorithm searches for a transformation vector with the highest fitness value. The obtained transformation vector gives the scale, translation and orientation of the model in the image.

A computer-aided diagnosis system for lung nodule detection in chest radiographs using a two-stage classification method based on radial gradient and template matching

In this paper we propose a scheme for automated detection of lung nodules in chest radiographs. The proposed scheme first segments lungs in a chest image using an active shape model. Next, the scheme detects initial nodule candidates by using a method previously reported by the authors. After that, the proposed scheme classifies nodule candidates into nodules and false positives by using a two-stage classification method proposed in this paper. For performance evaluation of the proposed nodule detection scheme, we made experiments using 125 images with nodules in the JSRT database which is a public database. We created 40 data sets by 40 randomized selection of 80 training images and 45 test images from the 125 images. As the result of experiments using these 40 data sets, the proposed scheme gave 6.6, 7.6, and 9.1 false positives per image for sensitivity values of 60.1, 64.1, and 69.7% on the average of 40 data sets. The time needed by the proposed scheme was 8.2 seconds per image on the average of 40 data sets using 3.3GHz Intel PC.