Canonical correlation analysis (CCA) is a popular method that has been extensively used in feature learning. In nature, the objective function of CCA is equivalent to minimizing the distance of the paired data, and L2-norm is used as the distance metric. We know that L2-norm-based objective function will emphasize the large distance pairs and de-emphasizes the small distance pairs. To alleviate the aforementioned problems of CCA, we propose an approach named CCA based on L1-norm minimization (CCA-L1) for feature learning. To optimize the objective function, we develop an algorithm that can get a global optimized value. To maintain the distribution and the nonlinear characteristic respectively, we proposed two extensions of CCA-L1. Further, all of the aforementioned three proposed algorithms are extended to deal with multifeature data. The experimental results on an artificial dataset, real-world crop leaf disease dataset, ORL face dataset, and PIE face dataset show that our methods outperform traditional CCA and its variants.
Calibration of the x-ray apparatus is necessary for many applications. Most of the literature on the calibration of x-ray apparatus seem to ignore the imaging deformation. Our main concern is how to apply Tsai's nonlinear camera calibration technique to the calibration of a medical x-ray apparatus with deformation. In order to achieve this goal, two key problems should be solved: The first is how to calibrate some key intrinsic parameters, namely, the imaging center and sampling step of the detector, which are usually provided by the camera manufacturer but are unknown for the x-ray apparatus. The second is how to model the serious imaging deformation. Some practical schema are designed to solve these two problems, and the whole calibration procedure and experimental results are presented.
In this paper some properties of Foley-Sammon optimal discriminant vector (FSODV), by contrast with uncorrelated optimal discriminant vector (UODV), are discussed. Firstly the Fisher ratio of every FSODV must be not less than that of corresponding UODV and consequently sole FSODV will be superior to corresponding UODV. Secondly the correlation between feature components extracted by FSODV is an important factor. If high correlation is available between most of the feature components, the classification performance of FSODV will be remarkably inferior to UODV. However, if most of the feature components are only little correlative to each other, FSODV is comparative to UODV in classification.
Many existing methods for face detection using both the positive examples (faces) and negative examples (nonfaces). By learning only from the positive examples, a novel face detection algorithm is invented, which is made up of two parts of research works. The first one is a frontal-view upright face detection algorithm based on the well-known singular value feature (SVF) and Hidden Markov models (HMM). The algorithm couples the virtues of both the SVF and HMM and produces excellent detection results. Firstly, it is tested on the second part of a large face image library NUSTFDB603-II whose first part is used to train the HMM and where there are 954 face images of 96 persons. The detection rate is 98.32% while only one false alarm is reported. Then it is tested on a collect photo album and has detected the 85.1 percent of its 484 people, while 97 false alarms are also reported. The second part of our algorithm is the extension of the first one to rotation-invariant face detection. Several HMMs are employed simultaneously and the angle of the "face" image is obtained. Then the HMM for detecting the upright faces is employed to verify the faceness of the test pattern. This rotation-invariant algorithm is tested on another image set where there are 173 persons whose faces are rotated randomly. The detection rate is 72.2%, and 34 false alarms are reported.
This paper proposed a set of practical method for corner detection used for X-ray device calibration. Based on the square steel pattern, the original image is first segmented into several small squares, then edge detection is operated to each small square, after this, Hough transform is used to line approximation and cross points of lines are solved, these cross points are just the corner needed. The experimental results show the proposed method is feasible and robust.
In this paper, we propose a set of effective algorithms to automatically detect the lung cancer cells in the cytological color image of examines' sputum smears. To increase the stability and efficiency of the detection of the cancer cells, a hierarchical processing architecture is adopted for the segmentation and recognition. For segmentation, RGB space and Lab space are combined to segment cell. By this method, both the nucleus and cytoplasm of cancer cells can be separated from background. Then, the candidate cancer cells are selected using some morphological features of nuclei, the purpose of this step is to pick out most of non cancer cells and leave a few doubtful cells for further verification, therefore improve the efficiency of the whole recognition process. As the last step, all the candidate cell, some statistic parameters in different color space are calculated, which are used as features for recognition. Experiment results are given.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.