Jiann-Shu Lee

Multiscale corner detection by using wavelet transform

A multiscale corner detection algorithm based on the wavelet transform of contour orientation is proposed. It can utilize both the information of the local extrema and modulus of transform results to detect corners and arcs effectively. The ramp-width of contour orientation profile, which can be computed using the transformed modulus of two scales, reveals the difference between corner and arc and is utilized in the determination of corner points. The experimental results have shown that the detector is more effective than both the single- and multiple-scale detectors. They also demonstrate that the detector is insensitive to boundary noise. In addition, the proposed method is more efficient than the other multiscale corner detector because it operates on fewer number of scales, which can be implemented by a fast transform algorithm.

Wavelet transformation for gray-level corner detection

Abstract Corners are very attractive features for many applications in computer vision. In this paper, a new gray-level corner detection algorithm based on the wavelet transform is presented. The wavelet transform is used because the evolution across scales of its magnitudes and orientations can be used to characterize localized signals like edges and corners. Most conventional corner detectors detect corners based on the edge detection information. However, these edge detectors perform poorly at corners, adversely affecting their overall performance. To overcome this drawback, we first propose a new edge detector based on the ratio of the inter-scale wavelet transform modulus. This edge detector can correctly detect edges at the corner positions, making accurate corner detection possible. To reduce the number of points required to be processed, we apply the non-minima suppression scheme to the edge image and extract the minima image. Based on the orientation variance, these non-corner edge points are eliminated. In order to locate the corner points, we propose a new corner indicator based on the scale invariant property of the corner orientations. By examining the corner indicator the corner points can be located accurately, as shown by experiments with our algorithm. In addition, since wavelet transform possesses the smoothing effect inherently, our algorithm is insensitive to noise contamination as well.

Wavelet based corner detection

Abstract A non-parametric algorithm for detecting and locating corners of planar curves is proposed. The algorithm is based on the multiscale wavelet transform of the orientation of the curve which can effectively utilize both the information of local extrema positions and magnitudes of the transform results. The corner candidates can then be selected easily based on this information. According to the angle preserving concept, intrinsic ratios of several corner models have been derived and used to evaluate the corner candidates. The corner angles can also be obtained during these processes. To make the evaluation process robust a masking algorithm is proposed. Experiments depict that our detector is more effective than the single scale corner detectors, while is more efficient than the multiscale corner detector by Rattarangsi and Chin ( Proc. Int. Conf. Pattern Recognition , pp. 923–930 (1990)).

Endocardial boundary detection using a neural network

Abstract Echocardiography has been widely used as a real-time non-invasive clinical tool to diagnose cardiac functions. Due to the poor quality and inherent ambiguity in echocardiograms, it is difficult to detect the myocardial boundaries of the left ventricle. Many existing methods are semi-automatic and detect cardial boundaries by serial computation which is too slow to be practical in real applications. In this paper, a new method for detecting the endocardial boundary by using a Hopfield neural network is proposed. Taking advantage of parallel computation and energy convergence capability in the Hopfield network, this method is faster and more stable for the detection of the endocardial border. Moreover, neither manual operations nor a priori assumptions are needed in this method. Experiments on several LV echocardiograms and clinical validation have shown the effectiveness of our method in these patient studies.

A Visual Context-Awareness-Based Sleeping-Respiration Measurement System

Due to the rapid growth of the elderly population, improving specific aspects of elderly healthcare has become more important. Sleeping care systems for the elderly are rare. In this paper, we propose a visual context-aware-based sleeping-respiration measurement system that measures the respiration information of elderly sleepers. Accurate respiration measurement requires considering all possible contexts for the sleeping person. The proposed system consists of a body-motion-context-detection subsystem, a respiration-context-detection subsystem, and a fast motion-vector-estimation-based respiration measurement subsystem. The system yielded accurate respiratory measurements for our study population.

The Adult Image Identification Based on Online Sampling

This paper presents a new adult image identification system. We propose the online skin tone sampling mechanism based on face detection. The object detection based on machine learning approach is used to locate the face region in the target image. According to the face color information, the skin chromatic distribution is extracted to detect the skin objects in the target image. The texture feature, coarseness, is also utilized to acquire the more accurate skin regions. Then, the back propagation neural network is used to integrate several low-level and reliable geometrical constraints into a classifier to inspect the skin regions further. Finally, the mug shot exclusion procedure is applied to promote the system performance. The experimental results show our method is satisfactory for adult image detection.

Occluded objects recognition using multiscale features and hopfield neural network

Abstract A new method to recognize partially visible two-dimensional objects by means of multiscale features and Hopfield neural network was proposed. The Hopfield network was employed to perform global feature matching. Since the network only guarantees to converge to a local optimal state, the matching results heavily depend on the initial network state determined by the extracted features. To acquire more satisfactory initial matching results, a new feature vector was developed which consists of the multiscale evolution of the extremal position and magnitude of the wavelet transformed contour orientation. These features can even be used to discriminate dominant points, hence good initial states can be obtained. The good initiation enables our proposed method to recognize objects even heavily occluded, that cannot be achieved by using the Nasrabadi-Lis method. In addition, to make the matching results more insensitive to the threshold value selection of the network, we replace the step-like thresholding function by a ramp-like one. Experimental results have shown that our method is effective even for noisy occluded objects.

Boundary-based corner detection using wavelet transform

A novel corner detection algorithm based on the wavelet transform is presented. The algorithm detects corners by applying the shape information of the orientation profile of the corner. The shape information is extracted by using the wavelet transform. The conducted experiments have shown that our algorithm is more effective than the conventional corner detection algorithms. Compared with the traditional multiscale corner detection algorithms, our algorithm is computationally simpler because we have to compute the wavelet transform for only one or two scales.<<ETX>>

Color face image enhancement using adaptive singular value decomposition in fourier domain for face recognition

Face recognition is still a challenging problem because of large intra-class variability, small inter-class variability, and the presence of lighting variation. To deal with these difficulties, an illumination compensation method, adaptive singular value decomposition in the two-dimensional discrete Fourier domain (ASVDF) and an efficient brightness detector for lighting detection, for face image enhancement are proposed in this paper. The proposed enhancement algorithm involves three steps: In the first step, uniform lighting is rapidly distinguished from lateral lighting in the image by using the brightness detector, which is based on the percentage ratio of pixels among the three RGB color channels. ASVDF is then globally performed for the uniform lighting image, whereas ASVDF is applied block-by-block for the lateral lighting image. In addition, to reduce computing time, a region-based ASVDF method is introduced; the ASVDF method is applied to four regions of the face image. Experimental results for the CMU-PIE, Color FERET, and FEI face databases show that the method considerably improves the quality of face images, even lateral lighting, thereby improving the accuracy of face recognition substantially. Adaptive singular value decomposition in the Fourier domain is proposed for face recognition.Self-adapted illumination compensation is devised to overcome lighting variation.Experimental results are demonstrated on CMU, FERET, and FEI databases to verify the effectiveness.

The nude image identification with adaptive skin chromatic distribution matching scheme

This paper presents a new nude image identification system. We propose an adaptive chroma-distribution matching scheme based on face detection to on-line determine the images skin chromatic distribution such that it can tolerate the color deviation coming from special lighting without increasing false alarm. The object detection based on machine learning approach is used to locate the face region in the test image. According to the color information of face, the matched skin chromatic distribution is selected to detect the skin objects in the test image. The texture feature, namely coarseness, is used to acquire accurate skin segmentation. The low-level but reliable geometrical constraints and the mug shot exclusion procedure are employed to further examine the skin regions. Experimental results that the overall detection rate is 86.3% show our method can achieve satisfactory performance for detecting nude images under special lighting conditions.