In this paper, we propose a novel algorithm to recover a sharp image from its corrupted form by deconvolution. The algorithm learns the deconvolution process. This is achieved by learning the deconvolution filter kernels for the set of learnt basic pixel patterns. The algorithm consists of the offline learning and online filtering stages. In the one-time offline learning stage, the algorithm learns the dictionary of various local characteristics of the pixel patch as the basic pixel patterns from a huge number of natural images in the training database. Later, the deconvolution filter coefficients for each pixel pattern is optimized by using the source and the corrupted image pairs in the training database. In the online stage, the algorithm only needs to find the nearest matching pixel pattern in the dictionary for each pixel and filter it using the filter optimized for the corresponding pixel pattern. Experimental results on natural images show that our method achieves the state-of-art result on an image deblurring. The proposed approach can be applied to recover a sharp image for applications such as camera, HD/UHD TV, document scanning systems etc.
In this paper we present a technique to accurately build a 3D hyperspectral image cube from a 2D imager
overlaid with a wedge filter with up to hundreds of spectral bands, providing time-multiplexed data through
scanning. The correctness of the spectral curve of each pixel in the physical scene, being the combination of
its spectral information captured over different time stamps, is directly related to the alignment accuracy and
scanning sensitivity. To overcome the accumulated alignment errors from scanning inaccuracies, frequency-
dependent scaling from lens, spectral band separations and the imager’s spectral filter technology limitations,
we have designed a new image alignment algorithm based on Random Sample Consensus (RANSAC) model
fitting. It estimates many mechanical and optical system model parameters with image feature matching over
the spectral bands, ensuring high immunity against the spectral reflectance variations, noise, motion-blur, blur
etc. The estimated system model parameters are used to align the images captured over different bands in the
3D hypercube, reducing the average alignment error to 0.5 pixels, much below the alignment error obtained
with state-of-the-art techniques. The image feature correspondences between the images in different bands of
the same object are consistently produced, resulting in a hardware-software co-designed hyperspectral imager
system, conciliating high quality and correct spectral curve responses with low-cost.
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.