This paper provides a comprehensive analysis of exactly what visual information about the world is embedded within a single
image of an eye. It turns out that the cornea of an eye and a camera viewing the eye form a catadioptric imaging system. We
refer to this as a corneal imaging system. Unlike a typical catadioptric system, a corneal one is flexible in that the reflector
(cornea) is not rigidly attached to the camera. Using a geometric model of the cornea based on anatomical studies, its 3D
location and orientation can be estimated from a single image of the eye. Once this is done, a wide-angle view of the environment
of the person can be obtained from the image. In addition, we can compute the projection of the environment onto the retina
with its center aligned with the gaze direction. This foveated retinal image reveals what the person is looking at. We present
a detailed analysis of the characteristics of the corneal imaging system including field of view, resolution and locus of
viewpoints. When both eyes of a person are captured in an image, we have a stereo corneal imaging system. We analyze the epipolar
geometry of this stereo system and show how it can be used to compute 3D structure. The framework we present in this paper
for interpreting eye images is passive and non-invasive. It has direct implications for several fields including visual recognition,
human-machine interfaces, computer graphics and human affect studies.
This research was conducted while the first author was affiliated with Columbia University. A shorter version of this paper
appeared in (Nishino and Nayar, 2004). 相似文献
We present a new measure of image focus. It is based on wavelet transform of the image and is defined as a ratio of high-pass band and low-pass band norms. We show this measure is monotonic with respect to the degree of defocusation and sufficiently robust. We experimentally illustrate its performance on simulated as well as real data and compare it with existing focus measures (gray-level variance and energy of Laplacian). Finally, an application of the new measure in astronomical imaging is shown. 相似文献
A hybrid neural network is presented for the segmentation of ultrasound images.
Feature vectors are formed by the discrete cosine transform of pixel intensities in region of interest (ROI). The elements and the dimension of the feature vectors are determined by considering only two parameters: The amount of ignored coefficients, and the dimension of the ROI.
First-layer-nodes of the proposed hybrid network represent hyperspheres (HSs) in the feature space. Feature space is partitioned by intersecting these HSs to represent the distribution of classes. The locations and radii of the HSs are found by the genetic algorithms.
Restricted Coulomb energy (RCE) network, modified RCE network, multi-layer perceptron and the proposed hybrid neural network are examined comparatively for the segmentation of ultrasound images. 相似文献
Numerically optimized hard pulse sequences were developed providing a frequency selective response of the transverse and longitudinal magnetization for several applicationsin vivo. The sequences are based on pulse trains with binomial ratios of the pulse angles and constant time intervals between the pulses. These pulse trains were systematically optimized to obtain broad Larmor frequency ranges in which the magnetization is not markedly influenced by the pulse sequence. In addition, the sequences had to provide maximum transverse magnetization or complete inversion of the magnetization beside the suppression range. Such behaviour is needed for chemical shift selective imaging, pulsed magnetization transfer, and frequency selective spectroscopyin vivo. The phase of the magnetization response is shown and adequate rephasing conditions are discussed as well as the actual phase duration. Short optimized hard pulse sequences for water and fat-selective imaging provide lowT2-sensitivity of the resulting magnetization. The new optimized hard pulse sequences are suitable, if optimum suppression of signals at one single point in the Larmor frequency spectrum is less important than good suppression in a larger frequency range. 相似文献
Purpose: To evaluate the feasibility of MRI-based myocardial first-pass contrast perfusion imaging with a multi-shot echo planar imaging (EPI) technique.
Subjects and methods: A non-sequential (ECG-triggered) gradient echo two-shot EPI acquisition strategy capable of covering the entire heart in contiguous 10-mm sections every two cardiac cycles with an in-plane resolution of 1.56 × 1.56 mm was implemented on a 1.5-T Signa Advantage Scanner equipped with prototype hardware for non-resonant EPI in the transverse plane. The heart of a single volunteer was studied prior to and following the intravenous bolus application of a paramagnetic contrast agent (Gd-DOTA, 0.2 mmol/kg).
Results: Twelve contiguous transaxial 10-mm EPI images were obtained every two RR intervals for a total of 40 s. The myocardial contrast perfusion study was technically adequate. Contrast caused a signal loss of 87% in the right and 67% in the left ventricle and 59% in the myocardium.
Conclusion: First-pass myocardial perfusion imaging with a gradient echo, two-shot echo planar imaging strategy is feasible.This work has been supported in part by SNF grant 32-2549.88 and KWF grant 2194.1. 相似文献
To investigate the application of a mini-coil surface system for high-resolution MRI, 60 volunteers were examined in a 1.5-T whole-body scanner. Two replaceable probe heads were available: a circular 2.5-cm coil and a quadratic 5-cm coil, both of which were placed directly on the skin. The skin layers, Achilles tendon and finger joints were examined with the 2.5-cm coil and a FOV of 25 × 25 mm2. A matrix of 256 × 256 pixels resulted in a pixel size of 0.098 × 0.098 mm2. For imaging of the carpal tunnel, the 5-cm coil was used in transverse orientation. The FOV was 50 × 50 mm2 so that a matrix of 256 × 256 pixels led to a pixel size of 0.195 × 0.195 mm2. The resulting spatial resolution permitted visualization of the epidermis, dermis and subcutis, resulting in clear definition of anatomical detail of the musculoskeletal system. MRI measurement of skin-layer thickness did not correlate with histometric data (p<0.05). This discrepancy was due in part to shrinkage of the tumor specimen on histologic preparation. Other causes include the motion artifacts and the limited accuracy of determining thickness on the MRI display unit. 相似文献