In order to detect kidney cancer automatically from abdominal UCT (unenhanced CT) or CECT (contrastenhanced CT) images at an early stage, a promising approach is to use deep learning techniques with convolutional neural networks (CNNs). However, there still seem to be several challenges in detection of kidney cancer. For example, it is necessary to cope with the wide variety of abdominal CT images. In this paper, as an empirical study, we attempt to construct a CNN that detects kidney cancer well from abdominal CT images, with a special focus on how visual explanations produced by Gradient-weighted Class Activation Mapping (Grad-CAM) help us to construct such a CNN.
This paper presents a robust object tracking method under pose variation. In practical environment, illumination and pose of objects are changed dynamically. Therefore, the robustness to them is required for practical applications. However, it is difficult to be robust to various changes by only one tracking model. Therefore, the robustness to slight variations and the easiness of model update are required. For this purpose, Kernel Principal Component Analysis (KPCA) of local parts is used. KPCA of local parts is proposed for the purpose of pose independent object recognition. Training of this method is performed by using local parts cropped from only one or two object images. This is good property for tracking because only one target image is given in practical applications. In addition, the model (subspace) of this method can be updated easily by solving an eigen value problem. However, simple update rule that only the tracked region is used to update the model for next frame may propagate the error to the following frames. Therefore, the first given image which is a unique supervised sample should be used effectively. To reduce the influence of error propagation, the first given image and tracked region in t-th frame are used for constructing the subspace. Performance of the proposed method is evaluated by using the test face sequence captured under pose, scaling and illumination variations. Effectiveness of the proposed method is shown by the comparison with template matching with update. In addition, adaptive update rule using similarity with current subspace is also proposed. Effectiveness of adaptive update rule is shown by experiment.
This paper presents an object detection method using independent local feature extractor. In general, it can be considered that objects are the combination of characteristic parts. Therefore, if local parts specialized for recognition target are obtained automatically from training samples, it is expected that good object detector is developed. For this purpose, we use Independent Component Analysis (ICA) which decomposes a signal into independent elementary signals. The basis vectors obtained by ICA are used as independent local feature extractors specified for detection target. The feature extractors are applied to candidate region, and their outputs are used in classification. However, the extracted features are independent local features. Therefore the relative information between neighboring positions of independent features may be more effective for object detection than simple independent features. To extract the relative information, higher order local autocorrelation features are used. To classify detection target and non-target, we use Support Vector Machine which is known as binary classifier. The proposed method is applied to car detection problem. Superior results are obtained by comparison with Principal Component Analysis.
This paper presents scale invariant face detection and classification methods which use spectral features extracted from Log-Polar image. Scale changes of a face in an image are represented as shift along the vertical axis in Log-Polar image. In order to make them robust to the scale changes of faces, spectral features are extracted from the each row of the Log-Polar image. Autocorrelations, Fourier power spectrum, and PARCOR coefficients are used as spectral features. Then these features are combined with simple classification methods based on the Linear Discriminant Analysis to realize scale invariant face detection and classification. The effectiveness of the proposed face detection method is confirmed by the experiment using the face images which are captured under the different scales, backgrounds, illuminations, and dates. We have also performed the experiments to evaluate the proposed face classification method using 2800 face images with 7 scales under 2 different backgrounds.
The face recognition, as one of the pattern recognition, includes various essence such as the representation and the extraction of the required features, the classification based on the obtained features and the detecting specified regions etc. Previously, we presented the scale and the rotation invariant face recognition method based on both Higher-Order Local Autocorrelation features of log-polar image and linear discriminant analysis for 'face' and 'not face' classification. In this face recognition method, the searching for the 'face' region was performed randomly or sequentially on the image. Therefore its searching performance was not satisfiable. In this paper, we present a method to narrow down the search space by dynamically using the information obtained at the previous search point through constructing the multilevel dynamic attention map, which is constructed based on the Ising dynamics and the renormalization group method.
In this paper, a multilevel Ising search method for human face detection is proposed to speed up the search. In order to utilize the information obtained from the previous searched points. Ising model is adopted to represent the candidates of `face' positions and is combined with the scale invariant human face detection method. In the face detection, the distance from the mean vector of `face' class in discriminant space represents the likelihood of face. By integrating the measured distance into the energy function of Ising model as the external magnetic field, the search space is narrowed down effectively (the candidates of `face' are reduced). By incorporating color information of face region in the external magnetic field, the `face' candidates can be reduced further. In the multilevel Ising search, face candidates (spins) with different resolutions are represented in a Pyramidal structure and the coarse-to-fine strategy is taken. We demonstrate that the proposed multilevel Ising search method can effectively reduce the search space and can detect human face correctly.
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.