Images obtained by an aberration-free system are defocused blur due to motion in depth and/or zooming. The
precondition of restoring the degraded image is to estimate point spread function (PSF) of the imaging system as
precisely as possible. But it is difficult to identify the analytic model of PSF precisely due to the complexity of the
degradation process. Inspired by the similarity between the quantum process and imaging process in the probability and
statistics fields, one reformed multilayer quantum neural network (QNN) is proposed to estimate PSF of the defocus
blurred image. Different from the conventional artificial neural network (ANN), an improved quantum neuron model is
used in the hidden layer instead, which introduces a 2-bit controlled NOT quantum gate to control output and adopts 2
texture and edge features as the input vectors. The supervised back-propagation learning rule is adopted to train network
based on training sets from the historical images. Test results show that this method owns excellent features of high
precision and strong generalization ability.
In this paper, an approach based on the quantum neural network is investigated to guide the process of selecting an
optimal estimation of Gaussian degraded parameter. In fact, we first formulate the nonlinear problem by maximum
likelihood estimation. Then we modify and apply the quantum neural network algorithm, which combines the advantages
of both quantum computing and neural computing, to solve the optimal estimation problem. The new algorithm does not
suffer from the morass of selecting good initial values and being stuck into local optimum as usually accompanied with
the conventional techniques. The simulation results indicate the soundness of the new method.
We proposed a novel method to extract and reduce Category Specific SIFT Descriptor (CSSD). Our approach is based
on two facts. One is that in many images there are always more than two different objects and this brings on ambiguity
of the categories which they should belong to. The other is that the number of SIFT features for one image often varies
from tens to thousands, matching these SIFT features of two arbitrary images brings high computational costs. As for the
first fact, those category specific SIFT features hide among the sum of SIFT information, we aim to filter out the
contributive SFIT information to category recognition by clustering all. With respect to the second fact, the SIFT clusters
can instruct to reduce each image SIFT features by keeping high occurrence frequency ones. So the more precious and
smaller SIFT features depending on its category specific features can be obtained. Another main highlight of our
approach is the sensible use of affinity propagation to address the definition of clustering category K more objectively.
Extensive experiments shows that the RCSSD (Reduced CSSD) obtained by affinity propagation clustering outperforms
the original SIFT descriptor and RCSSD by using K-means approach.
Proc. SPIE. 7383, International Symposium on Photoelectronic Detection and Imaging 2009: Advances in Infrared Imaging and Applications
KEYWORDS: Signal to noise ratio, Digital signal processing, Spatial frequencies, Image processing, Image restoration, Linear filtering, Image quality, Data processing, Algorithm development, Filtering (signal processing)
Up to now, there are a large number of image restoration algorithms developed and implemented on power-efficient
hardware platforms. In this paper a fast restoration approach based on edge-preserving is proposed and ported on a multi-
DSP platform. Firstly, classical Wiener filter is optimized and the blurred image is decomposed into two independent
parts in frequency domain: a shift-invariant part and a shift-variant part. Then the result is obtained by combining the two
parts. Secondly, parallel processing is adopted for the large volumes of data and the complex algorithms commonly
encountered. The algorithm mentioned above is realized on a parallel system with 4 DSPs because one DSP can not
afford such large amounts of data. In order to make full use of processors and memory, the timing and throughput is
designed carefully to guarantee the data processed in a pipeline manner. In a word, experiments show the algorithm has
higher performance than other methods with the same computational complexity and can achieve real-time processing.
As a fundamental image processing operation, a good denoising method should keep the original image information as
much as possible. However, most denoising methods may degrade or remove the fine details and texture of the original
image. In this paper, a force field method is adopted to transform the image pixels within a local window into a potential
energy surface and then to distinguish the image edges and the noises in this potential energy field. Afterwards, different
templates are used according to the judgment and the adaptive filter is applied to the local pixels respectively. This new
method has less computational complexity than the other algorithms of transform domain, which means it can be
implemented in a real-time processing system. Also the new method can preserve more image edges than the traditional
filters. Finally the performance of the proposed method is compared in this paper with other popular methods by using
evaluation criterion of SNR and SSIM(a measure of structural similarity). The results show that the proposed method is
reliable and especially helpful to preserve the image details.