Unit 4

IMAGE RESTORATION

IMAGE RESTORATION
4.1 INTRODUCTION:
Image restoration in an area deals with improving the appearance of an image. However, unlike enhancement, which is subjective, image restoration is objective, in the sense that restoration techniques tend to based on mathematical or probabilistic models of image degradation. Restoration attempts to reconstruct or recover an image that has been degraded by using a priori knowledge of the degradation phenomenon. Thus restoration techniques are oriented towards modeling the degradation and applying the inverse process in order to recover the original image.

4.2 IMAGE DEGRADATION MODEL:

A model of the Image Degradation / Restoration process

The fig. shows, the degradation process is modeled as a degradation function that, together with an addition noise term, operates on an input image f(x, y) to produce a degraded image g(x, y). given g(x, y) some knowledge about degradation function H, and some knowledge about the additive noise term (x, y) the objective of restoration is to obtain an estimation of the original image. The requirement is the estimate be as close as possible to the original input image and, in general, the more we know about H and , the closer will be to f(x, y). the approach used throughout most of this is based on various types of images restoration filters. If H is a linear, position invariant process, then the degradation image is given in the spatial domain by . Where is the spatial representation of the degradation function and as symbol * indicates spatial convolution. Convolution in the spatial domain is equal to multiplication in the frequency domain, the equivalent frequency domain representation. .. (4.1) Where the terms in capital letters are Fourier transforms of corresponding terms, H is identity operator. 4.2.1 INVERSE FILTERING: The simplest approach to restoration is direct inverse filtering, where an estimate, of the transform of the original image simply by dividing the transform of the degraded image G(u, v), by the degradation function. = (4.2) The divisions are between individual elements of the function where G(u, v) = H(u, v) * F(u, v) + N(u, v) substitute G(u, v) in eq.(1) i.e.,

Unit 4

IMAGE RESTORATION

= F(u, v) +

.. (4.3)

The expression tells us that even if we know the degradation function we cannot recover the underground image [ inverse fourier transform of F(u, v)] exactly because N(u, v) is a random function whose fourier transform is not known. If the degradation has zero or very small values then the ratio N(u, v) / H(u, v) could easily dominant the estimate . One approach to get around the zero or small value problem is to limit the filter frequency to the value near the origin. We know that H(0, 0) is equal to the average value of h(x, y) and that is usually highest value of H(u, v) in the frequency domain. Thus by limiting the analysis to frequencies near the origin, we reduce the probability of encountering zero values. Note: Consider the Gaussian function at high frequency H(u, v) is low. Hence 1/H(u, v) will be large i.e., the noise is more in order to over come this problem we are not considering the high frequency components we are setting the range in which we can have effective picture.

frequency 4.2.2 MINIMUM MEAN SQUARE ERROR (WIENER) FILTERING: Inverse filtering approach makes no explicit provision for handling noise. The method is founded on considering images and noise as random process and the objective is to find an estimate of the uncorrupted image f such that the mean square error between them is minimized. This error measures is given by (4.4) Where E{.} is the expected value of argument. It is assumed that the noise and the image or uncorrelated that one or other has zero mean; and that the grey levels in the estimate are a linear function of the levels in the error function in eq.1 is gain in the frequency domain by the expression [ =[
| |

] ]

| | | |

=[

. (4.5)

When we used the fact that the product of a complete quantity with its conjugate is equal to the magnitude of complex quantity squared. This result is known as Wiener filter. The terms inside the brackets are also called as minimum mean square error filter or the least square error filter. Wiener filter does not have the same problem as the inverse filter with zeros in degradation function unless both and (u,v) are zero for the same values (s) of u and v. =degradation function =complex conjugate H(u,v)
2

Unit 4

IMAGE RESTORATION

| =H*(u,v)H(u,v) =| =| | =power spectrum of the noise | =power spectrum of the un-degraded image

As before, is the transform of the degradation function and G(u,v) is the transform of the degraded image. The restored image in the spatial domain is given by the inverse fourier of the frequency-domain estimated . Note that if the noise is zero, then the noise power spectrum vanishes and the wiener filter reduces to the inverse filter. When dealing | is a constant, which simplifies things with spectrally white noise, the spectrum | considerably. However, the power spectrum of the un-degraded image seldom is known. An approach used frequently when these quantities are not known or cannot be estimated is to appropriate by the expression [ | | | | ]

Where K is a specified constant. 4.2.3 CONSTRAINED LEAST SQUARE FILTERING: Power spectra of the un-degraded image and noise must be known. A constant estimate of the ratio of power spectra is not always a suitable solution. Knowledge of only mean and variance of the noise is required. These parameters usually can be calculated from a given degraded image. Another difference is that the wiener filter is based on minimizing a statistical criterion and as such it is optimal in an average sense. The algorithm presented in this section has the notable future that yields an optimal result for each image to which it is applied. The choice of one algorithm over the other will almost always be determined (at least partially) by the perceived visual quality of the resulting images. By using the definition of convolution given we can express in vector matrix form, as follows; y=HF + (4.7) Suppose that g(x, y) is of size MxN. Then we can form the first N elements of the vector g by using the image elements in the first row of g(x, y), the next N elements from the second row, and so on. The resulting vector will have dimensions MN+1. These are also the dimensions of f and as these vectors are formed in the same manner. The matrix H then has dimensions MN x MN. Its elements are given by the elements of convoluting DOUBT. The restoration problem can be reduced to a simple matrix manipulation. Manipulating vectors and matrices of the large sizes is not a trivial. The problem is complicated further the fact H is highly sensitive to formulating the restoration problem in matrix form does facilitate elements of restoration techniques. One way to DOUBT to noise sensitive is to base optimality of restoration on a measure of smoothness, DOUBT second derivative of an image. The restoration must be constrained the parameters of the problems at hand. Thus to find the SOUBT of a criterion function C, defined as C= . (4.71) subject to the constraint. (4.72)

Where is the Euclidian vector norm and is the error of the un-degraded image. The laplacian operator . The frequency solution to this optimization problem is given by the expression [|
| | |

.. (4.73)
3

Unit 4

IMAGE RESTORATION

Where parameter that must be adjusted so that the constraint is satisfied P(u,v) is the fourier transform of the function. P(x, y) = [ ]

P(x, y) end other relevant DOUBT domain functions, must be properly with zeros prior to computing their transform in eq.4 reduces to inverse filtering if is zero. A procedure for computing by iteration is as follows. Define an residual vector as =g-H therefore from eq. and by implication is a function of , then also a function parameter. It can be increasing function of the gamma wish to adjust as shown that ( ) =

the constraint eq.2 is highly satisfied. Because Where a is accuracy factor ( ) is monotonic, finding the value of is not difficult one approach is to 1. Specify an initial value of . 2. Compute . 3. Stoop when is satisfied otherwise return to step 2 after increasing If or decreasing use a new value of in eq.3 to recompute the optimum estimate .