Some Methods To Obtain High-Resolution Image

This paper mainly introduce the development and research of methods to obtain high-resolution image,and describe several methods to obtain high-resolution image.To obtain high-resolution image is an image processing technology that from low-resolution image to high-resolution image.With the widely used of the image processing,to obtain high-resolution image has an important value as a basis image processing operation.For example,looking for mineral deposits by enlarging the satellite image and analysis the change of climate;enlarging CT image to make doctors diagnosis easier;enlarging reconnaissance photos to find valuable goal;enlarging digital camera photos to prominent figures and some scenes and so on.However,in real life,due to physical and hardware constraints,the images we acquired all have a certain precision,sampling rate can not be unlimited fine.If we want to get higher-resolution image,it can only be achieved through software. In image processing,a primary issue of the acquisition of high-resolution image is that the increase of the amount of recording image data.In other words,it needs more data.When adding data to image,according to its own characteristics and requirements of image magnification,we can take some different measures.But the premise is how to make the distortion of the lager image as small as possible.So learning the method of enlarging image have great significance.The methods usually used to obtain high-resolution image is based on smoothing and interpolation to reduce noise.Smooth mainly achieved by filter,common filters are:Gaussian filter, wiener filter and median filter.Interpolation is a process of regeneration of image data,generally it divided into two steps:first,choosing a fitting model for the original image;second,resampling the fitting model by the desired sampling rate. (1)Traditional interpolation;(2)Adaptive image magnification;(3)Image magnification based on partial differential equations;(4)Image magnification based on multiscale analysis with wavelet transform.In chapter 2,we introduce two image interpolation methods based on cubic convolution.Cubic convolution can be parameterized and then optimized either for general performance characteristics or for optimal fidelity over an image ensemble with specific characteristics.Traditionally,the cubic kernel has been derived in one dimension with one parameter and applied to two-dimensional images in a separable fashion.However, images typically are statistically non-separable.Reichenbach and Geng derived a non-separable,two-dimensional cubic kernel with two parameters(designated 2D2PCC) and showed that it yielded better image interpolation accuracy than separable cubic convolution.2D-3PCC is the most general two-dimensional,piecewise-cubie interpolator defined on[-2,2]×[-2,2]with constraints for biaxial symmetry,diagonal symmetry(which,with biaxial symmetry,provides 90°rotational symmetry), continuity,and smoothness.The second kernel,with five parameters(designated 2D-SPCC),relaxes the constraint of diagonal symmetry,based on the observation that many images have rotationally asymmetric statistical properties.In chapter 3,we introduce two adaptive image magnifications.First,we introduce the method of image interpolation using coons surface patch with shape control parameters.Piecewise bicubic Coons surface patch with adjustable parameters to modify the length and direction of tangent vector at some control points is constructed for a digital image.Boundary curves are first adjusted to fit the image edges.Then the U-V tangent vectors at interior control points and twist vectors at corner points of the surface patch are adjuster to fit the whole image.Such an approach can counteract the smoothing effect of traditional methods greatly and keep the images edges still sharp and smooth.Then we introduce an adaptive image magnification based on the hyperbolic tangent function.To deblur edge after image magnifying,an adaptive edge sharpness preserving image magnification is put forword,which can fit edges of any direction,gradient and amplitude,utilizing the gradient imgormation of the image and the properties of the hyperbolic tangent function having reviewing the character of a typical edge.The experiments have verified the effectivity of the algorithm in this thesis.In chapter 4,we first introduce the method of image zooming based on Thiele's rational interpolation.It is a method of image zooming based on Thiele type continued fractions interpolation function and Newton polynomials interpolation function is presented.Newton interpolation polynomials are constructed when the construction of Thiele's rational interpolation functions are not successful.Experimental results show that the method can improve quality of zoomed image effectively.Then we introduce the Super-Resolution Through Neighbor Embedding.The neighbor embedding algorithm of our method can be summarized as follows:1.For each patch X_t~q in image X_t:(a)Find the set N_q of K nearest neighbors in X_s.(b)Compute the reconstruction weights of the neighbors that minimize the error of reconstructing X_t~q.(c)Compute the high-resolution embedding Y_t~q using the appropriate highresolution features of the K nearest neighbors and the reconstruction weights. 2.Construct the target high-resolution image Y_t by enforcing local compatibility and smoothness constraints between adjacent patches obtained in step 1(c).Then we introduce a sampled texture prior for image super-resolution.Superresolution aims to produce a high-resolution image from a set of one or more lowresolution images by recovering or inventing plausible high-frequency image content. Typical approaches try to reconstruct a high-resolution image using the subpixel displacements of several low-resolution images,usually regularized by a generic smoothness prior over the high-resolution image space.Other methods use training data to learn low-to-high-resolution matches,and have been highly successful even in the single-input-image case.Here we present a domain-specific image prior in the form of a p.d.f,based upon sampled images,and show that for certain types of super-resolution problems,this sample-based prior gives a significant improvement over other common multiple-image super-resolution techniques.In the end,we introduce the existing methods for image data fusion are not quite satisfactory for object detection.To improve the resolution of target and suppress the detection noise of each sence,a new method for image data fusion based on wavelet transform is presented.By decomposing the image with wavelet transform, wavelet coefficients and approximation coefficients at different scales are obtained. We took those coefficients with larger absolute value in the multi-resolution images as the important wavelet coefficients and computed the weighted mean value of the approximation coeffidents.And the fused image can be obtained by using the inverse wavelet transform for the impor-rant wavelet coefficients and the weighted approximation coefficients.Experimental results show that the data fusion method based on wavelet transform is very effective and can be applied to wide research fields.