K. R. Castleman, Digital Image Processing. New Jersey: Prentice Hall, 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....2 IR is u(x) G(x Gamma )u( d (6) where G(x) is the incoherent point spread function and u is the intensity distribution of the radiation field. In the case of a circular aperture of diameter a in a narrow band incoherent light having center wavelength , the point spread function is ([16], Ch. 15, 29] Ch. 10) G(x) 2 J 1 ( 7) where J 1 (r) is the first order Bessel function of the first kind, r is the radial distance in the image plane and r 0 = a ; 8) z being the distance from the lens to the image plane. In case of a square aperture [ Gammaa; a] Theta [ Gammab; ....

....order Bessel function of the first kind, r is the radial distance in the image plane and r 0 = a ; 8) z being the distance from the lens to the image plane. In case of a square aperture [ Gammaa; a] Theta [ Gammab; b] G(x 1 ; x 2 ) 9) where x 01 = 2a , x 02 = 2b ([16], Ch. 15, 29] Ch. 10) In view of the previous discussion, we shall assume that the optical system is described by a convolution operator with a positive kernel G. Moreover, we shall assume that G is either rotationally invariant or has a separable representation like the one in (9) Let us ....

[Article contains additional citation context not shown here]

K.R. Castleman, Digital Image Processing. Prentice Hall, 1996.

....presented in [31] However, the face images presented in [3] show that only extreme illumination direction conditions produce significant shadows, where even humans have trouble recognizing faces. See also Section 4. 5 for experiments on the Weizmann Database [1] Histogram equalization [9, 22] is often used in an attempt to reduce the effects of varying illumination conditions [29, 41] Method Time (msec) PCA 11 DCT 6 Gabor 675 DCT mod2 8 Table 1: Average time taken per face window (results obtained using Pentium III 500 MHz, Linux 2.2.18) effect is that the number of feature ....

K. R. Castleman, Digital Image Processing, Prentice-Hall, USA, 1996.

....The observed signal or image g is just the convolution of this blurring function h with the true signal or image f . The deblurring problem is to recover f from the blurred function g given the blurring function h. This basic problem appears in many forms in signal and image processing [2, 5, 12, 14]. In practice, the observed signal or image g is of finite length (and width) and we use it to recover a finite section of f . Because of the convolution, g is not completely determined by f in the same domain where g is defined. More precisely, if a blurred signal g is defined on the interval [a, ....

K. Castleman, Digital Image Processing, Prentice--Hall, Englewood Cli#s, NJ, 1996.

....scaled as # ### # # ### ## # # . The resultant image # is thus of a standard deviation #. the MSE metric is then applied to # , after image # is normalized in the same manner. 4.1. 3 # # : Template Matching (TM) Template matching is a commonly used technique in pattern recognition [1]. It examines the cross correlation (or autocorrelation) sequences in order to determine if a testing image contains a template image. Consider two images without any object shift. The conventional cross correlation function of two images # and # is: ######## # However ## is sensitive to the ....

....of both. Let ## , # # , # # and ## be the discrete Fourier transforms of #, #, # and #, respectively. Suppose #### ## represents a difference function between # and # in the Fourier domain, i.e. #### #### # ### ## ##### ##. According to the shift theorem and addition theorem [1], we have: ## ### ######## #### # ### ######## #### # # ### ## ###### #### #### ### Let # # # .Wedefineatransfer function # ### as: # ### ### ### ##### ## ##### ## # which is equal to ## # ## when # # #. We measure the significance of ##### ######## ##, in terms its ....

K.R.Castleman. Digital Image Processing. Prentice Hall, 1996.

....research are established. c #2003 Yang s Scientific Research Institute, LLC. All rights reserved. 1. INTRODUCTION The objective of image restoration is to reconstruct the image from degraded one resulted from system errors and noises and so on. There are two ways to achieve such an objective [3, 4, 29]. One is to model the corrupted image degraded by motion, system distortion, and additive noises, whose statistic models are known. And the inverse process may be applied to restore the degraded images. Another is called image enhancement, that is, constructing digital filters to remove noises to ....

....y(X) w i (X) x i , 5) where w i (X) is an adjustable parameter related to X. Obviously if the following fact holds for w 1 (X) w d (X) w 1 (X) w d (X) 0, 0 z c 1 , 1, 0, 0) card . d x i then y( defined by (5) is median filter [3, 4, 24, 29]; If the constants are determined by some adaptive learning algorithms, 5) gives an adaptive filter [2 5, 11, 35, 36] If w 1 (X) w d (X) are defined by some fuzzy sets and fuzzy rules, y( is a fuzzy weighted filter [2, 9, 14 16, 22, 24, 37] Assume that Z( is median filter. The fuzzy ....

K.R. Castleman, "Digital image processing", Beijing: Tsinghua University Press, Prentice Hall, Inc., 1998.

....of y(nx, n2) and the deconvolution estimate (nx, n2) respectively. The T(fx, f2) in (8) is called the regularization term and is set appropriately during deconvolution [20] For example, using the signal to noise ratio to set T(fx f2) Ir(f,f)P) in (7) yields the Wiener deconvolution estimate [24]; X(fx,f) the F(f, f2) and X(f, f2) denote the respective Fourier transforms of (n, n2) and x(n, n2) The in (7) constitutes the frequency response of the so called deconvolution filter. P(f,f2) X,f2 Fourier based deconvolution suffers from the drawback that its estimates for images ....

K. R. Castleman, Digital Image Processing. New Jersey: Prentice Hall, 1996.

....from the information stored in the index structure and the e#ciency. The most common dimensionality reduction approach in the literature is based on transformations such as the Discrete Fourier Transform (DFT) 38] the Discrete Cosine Transform (DCT) 29] the Discrete Wavelet Transform (DWT) [11], and Karhunen Loeve Transformation (KLT) 28] Since these transforms are distance preserving,ifapointisin# neighborhood of the query point, it remains in its # neighborhood after the transformation. The high dimensional original feature vectors are transformed, and the resulting vectors are ....

K. R. Castleman. Digital Image Processing. Prentice-Hall, Inc., 1996.

....presented in [30] However, the face images presented in [3] show that only extreme illumination direction conditions produce significant shadows, where even humans have trouble recognizing faces. See also Section 4. 5 for experiments on the Weizmann Database [1] Histogram equalization [8, 21] is often used in an attempt to reduce the effects of varying illumination conditions [28, 40] Method Time (msec) PCA 11 DCT 6 Gabor 675 DCT mod2 8 Table 1: Average time taken per face window (results obtained using Pentium III 500 MHz, Linux 2.2.18) effect is that the number of feature ....

K. R. Castleman, Digital Image Processing, Prentice-Hall, USA, 1996.

....As can be observed in Figure 5.4, the first three 2D DCT coe#cients contain a significant amount of person dependent information; thus ignoring them (as in DCT mod) implies a reduction in performance. This is verified in Figure 5. 5 where the DCT mod features have Histogram equalization [24, 41] is often used in an attempt to reduce the e#ects of varying illumination conditions [69, 89] 1 3 6 10 15 21 28 0 5 10 15 20 25 30 Dimension of DCT feature vectors ( Figure 5.4: Performance for varying dimensionality of 2D DCT feature vectors 0 5 10 15 20 25 DCT DELTA ....

K. R. Castleman, Digital Image Processing, Prentice-Hall, USA, 1996.

....the histogram as a feature vector. Some of these properties are retained in the histogram data, but some others are unique to the entries of the feature vector described above. Chief among these properties are translation, rotation and reflection. The histogram is invariant to these operations [15]. Hence, the mean, variance, and skewness remain unchanged. Therefore, the statistical feature vector remains unchanged as well. However, when point operations are performed on an image its histogram is modified in a predictable way. As a matter of fact, if f is a mapping function that resembles ....

....point operations are performed on an image its histogram is modified in a predictable way. As a matter of fact, if f is a mapping function that resembles a linear point operation (PB=a. PA b) performed on an image. Then the modified image histogram is related to the original histogram such that [ 15] (6) where PB and PA are the resulting and original image pixels, respectively. Consider two special cases of point operations, namely, adding a constant, and taking the negative of the image. We ll discuss the direct influence of each of these operations on the traditional image retrieval ....

K. Castleman, Digital Image Processing, Prentice Hall, New Jersey, 1996.

....produce a more consistent spatial classification by considering neighboring classifications. objects in the image (or, for locally adaptive thresholding, within the sliding window) One application where thresholding has been successfully applied is in segmenting chromosomes in microscope images [24]. Distance based classification assigns each pixel to the closest object class. Each class has a representative feature vector often called the cluster center and each pixel is assigned to the class that minimizes the distance to the pixel s feature vector. As in thresholding, a feature ....

K. R. Castleman, Digital Image Processing. Englewood Cliffs, NJ: Prentice Hall, 1996.

....both rectus muscles [4] Figure 2) C) Pre filtering Pre filtering has been reported to improve the quality of image derivatives [11;19;120;165] for the Lucas and Kanade algorithms. We have looked at: ## Gaussian smoothing, performed by convolution with a Gaussian kernel in 2D time [29], and ## nonlinear diffusion filtering, as developed by Perona et al. 124] by recursively filtering all images in a sequence using nonlinear di ffusion. The parameters for the di ffusion equation were set to the values reported in [11] and [124] This filter favors intra tissue smoothing by ....

.... on both sides with a 3 D Gaussian W with standard deviation (of the associated probability function) # w : # WAV Wb (4) where, for x i = 1 m, x i ### ## 12 diag( m WW W # Wxxx# The optimum solution for V in (4 ) in a l east squares sense, is obtained using the pseudoinverse [29]: # VWAWb (5) where ## WA = pseudoinverse of # # WA . Provided 0 AWA , the pseudoinverse is identical to the least squares inverse AWA A . It can be found efficiently in closed fo rm: 2 2 ( x x y xz T xy y yz xz yz z WI WII WII WII WI WII WII WII WI ## ## ## ## ## ....

[Article contains additional citation context not shown here]

Castleman, K. R., Digital Image Processing, Upper Saddle River, NJ: Prentice Hall, 1996.

....modified to be nearly balanced, then the information leaking becomes practically impossible (please refer to Fig. 2 of [13] Method 2. Introducing compression mechanism: after C i is obtained, compress it with any lossless entropy compression algorithm (such as Hu#man compression algorithm [27]) to cancel the information redundancy (i.e. to make the distribution of C i nearly balanced) and then mask the compressed C i with f be (x 0 ) Here, please note that the smaller C i is, the larger the occurrence probability will be, and the smaller the length of the compressed C i will ....

K. R. Castleman, Digital Image Processing, Prentice Hall Inc., New York, 1996.

.... Signal Processing 82 (2002) 749 757 SIGNAL PROCESSING www.elsevier.com locate sigpro Continuous wavelet transform with arbitrary scales and O(N) complexity Arrate Mufioz , Raphaal Ertl6, Michael Unser Del)artment of Microtechnolo,q) Biomedical lma,qing Groul) DMT, Sn,isx Federal Institute Technolo,qy Lausanne, ....

....is a 2 D description of the signal with respect to time b and scale a. The scale a is inversely proportional to the central frequency of the rescaled wavelet (x) x a) which is typically a bandpass function; b represents the time location at which we analyze the Corrcsponding author. Tcl. 41 21 693 5142; fax: 41 21 69 3701. E mail addresses: mTate.munozepfi.ch (A. Mufioz) raphael.ertle compaq.com (R. Ertlb) michael.unser epfi.ch (M. Unser) signal. The larger the scale a, the wider the analyzing function , and hence smaller the corresponding analyzed frequency. The output ....

[Article contains additional citation context not shown here]

K.R. Castleman, Digital Image Processing, Prentice-Hall, Englewood Cliffs, N J, USA, 1996.

....It means the removal or reduction of degradations in the concerned image, which includes the blurring and noise that are introduced while the digital image was obtained. Generally noise sources are divided into three categories, electronic noise, photoelectronic noise and film grain noise [8]. The majority of research activities have been devoted to modeling noise from these sources. In most cases, the noise can be modeled by a zero mean white Gaussian process. On some occasions, a non Gaussian process [1] is used as a more accurate characterization of the noise. Because image ....

K R. Castleman, Digital Image Processing, Prentice Hall, Inc., 1996.

....1 Introduction Uncompressed images require considerable storage capacity and very high bandwidth in transfer. In order to manage large image data objects eciently, these objects need to be compressed to reduce the le size. Compression tries to eliminate redundancies in the pattern of data [1] [6] Many studies have shown that users at a computer have a patience factor ranging from two to four seconds. Even though network speeds have been increasing consistently, very large image objects can take as much as a couple seconds to transmit. Given just a few seconds to retrieve, transmit, ....

....whose length is as small as possible. There are several typical transformations employed in image compression, such as the Fourier Transformation and the Wavelet Transformation. Besides transformation, most of the image compression algorithms also involve in some kind of quantization and encoding [1]. Our algorithms of background dilution apply to this class of compression algorithms. In this paper we use JPEG as an example to show how to modify JPEG to implement background dilution. JPEG is a sophisticated and exible standard for compressing and storing still images. JPEG de nes both ....

Kenneth R. Castleman. Digital Image Processing. Prentice Hall, New Jersey, 1996.

....scene has higher uncertainty than a smooth image, as will be shown in the following example. It is usually denoted as H(z) in bits per pixel unit of measurement. H(z) defines the average amount of information (in r ary units per symbol) obtained by observing a single source output [Gonzalez92][Castleman96]. If 2 is chosen as the base for logarithm, the units of entropy are bit per symbol. As the magnitude of the signal increases, more uncertainty and therefore more information are associated with the source data. Entropy is defined as 34 = J j j j a P a P z H 1 ) log ) 3.3) ....

....histograms of any selected subimage from the reference and the input data. Dissimilar histograms would indicate an anomaly and the corresponding subimage could be rejected to avoid misregistration. Histogram Matching Histogram matching has already been used in many image processing applications [Castleman96]. We have experimented with three matching methods: 1) The matched sample t test [Wadsworth98] which can be expressed as follows. N S D t D = 3.9) where D is a mean of differences between two histograms of the subimages, SD is a standard deviation of the differences, and N is the number of ....

K. R. Castleman, Digital Image Processing, Prentice-Hall, Inc., 1996.

....poor overlap between image neighborhood histograms; for example, a light A will be of little use when processing a dark B. As a preconditioning step, we would like to discover a luminance transformation that brings the histograms into correspondence. One standard approach is histogram matching [9]; however, we find that it uses non smooth mappings with undesirable side effects. Our approach, instead, is to apply a linear map that matches the means and variances of the luminance distributions. More concretely, if Y (p) is the luminance of a pixel in image A, then we remap it as Y (p) # ....

Kenneth Castleman. Digital Image Processing. Prentice-Hall, 1996.

....of the application. One major difficulty regards the selection of suitable features to be used in the classification, since there is no systematic and effective methodology for this task. Indeed, feature selection is nearly always carried out based on heuristics and trial and error procedures [Castleman (1996)] Furthermore, different combinations of possible feature vectors and pattern classifiers (since many classifiers can be adopted) seldom lead to the same classification results, indicating the criticality of a suitable choice of a feature vector pattern classifier scheme. There are too many ....

....to minimize dispersion within each cluster while maximizing the distance between clusters. In this approach, the expert classification underlying case 2 of Section 2 is simply disregarded, and an exclusively data driven fitness function is used instead. An example is the class separation distance [Castleman (1996)] between two clusters (denoted as 1 and 2) defined as D 1,2 = 1 2 (s 1 2 s 2 2 ) 1 2 , where 1 and 2 are the means of the clusters 1 and 2, respectively, while s 1 and s 2 are the respective standard deviations. On the other hand, if case 1 of Section 2 is being considered ....

[Article contains additional citation context not shown here]

K.R. Castleman, Digital Image Processing, Prentice-Hall, Englewood Cliffs, NJ, 1996.

....designing a shape based image retrieval system. If we want a system to distinguish objects of different types, we must first decide which parameters of the objects will be measured. The particular parameters that are measured are called the features. Good features should have four characteristics [5]: a) Discrimination. Features should take on significantly different values for objects belonging to different classes. b) Reliability. Features should take on similar values for all objects of the same class. c) Independence. The various features used should be uncorrelated with each other. ....

....be found in O(WH) where W and H are the width and the height of a frame buffer, respectively. Therefore, the computational complexity is O(WH) and the storage complexity of this representation method is O(1) This method is indexable. 3.1. 2 Rectangularity Idea Given an object, rectangularity R [5, 59] is defined as R = A AR : 3.3) where A is the area of the object and AR is the area of its minimum enclosing rectangle. Discussion It represents how well an object fills its minimum enclosing rectangle. This parameter takes on a maximum value of 1.0 for rectangular objects. The ....

K.R. Castleman. Digital Image Processing. Prentice-Hall, INC., Englewood Cliffs, N.J., 1979.

No context found.

K. R. Castleman, Digital image processing, Prentice Hall, 1996.

No context found.

K. R. Castleman, Digital Image Processing. New Jersey: Prentice Hall, 1996.

No context found.

K. R. Castleman, Digital Image Processing. New Jersey: Prentice Hall, 1996.

No context found.

K. R. Castleman, Digital Image Processing, Prentice Hall, New Jersey, 1996.

No context found.

K. Castleman, Digital Image Processing, Prentice-Hall, New Jersey, 1996.

No context found.

K. Castleman, Digital Image Processing, Prentice-Hall, New Jersey, 1996.

No context found.

Castleman, K.R. (1979) Digital Image Processing. Prentice Hall, Englewood Cliffs, NJ.

No context found.

Castleman, K. R. (1979) Digital Image Processing, Englewood Cliffs, NJ: Prentice--Hall.

No context found.

Castleman KR. Digital Image Processing. Englewood Cliffs, NJ: Prentice-Hall, 1996.

No context found.

K.R. Castleman, Digital Image Processing, Prentice-Hall, Englewood Cliffs, NJ, 1996.

No context found.

K. R. Castleman, Digital Image Processing, Prentice Hall Inc., New York, 1996. 13

No context found.

Castleman, Kenneth R. Digital Image Processing. Prentice Hall, 1979.

No context found.

K. R. Castleman, Digital Image Processing. New Jersey: Prentice-Hall, Englewood Cliffs, 1996.

No context found.

K.R. Castleman. Digital Image Processing. Prentice Hall, Upper Saddle River, NJ, 1996. 11.4.1, 11.4.2

No context found.

K.R. Castleman. Digital image processing. Prentice-Hall, Englewood Cliffs, NJ, USA, 1996.

No context found.

Castleman, K. Digital Image Processing, Prentice-Hall, Englewood Cliffs, NJ, 1996.

No context found.

K.R. Castleman, Digital Image Processing. Prentice-Hall 1996.

No context found.

K. Castleman, Digital Image Processing, Prentice-Hall, Upper Saddle River, NJ, 1996.

No context found.

K.R. Castleman, Digital Image Processing. Prentice-Hall 1996.

No context found.

Castleman, K. R., "Digital Image Processing," Prentice Hall Inc., Inglewood, New Jersey, 1996.

No context found.

Castleman, K.R., Digital Image Processing. Second ed. 1996, Englewood Cliffs, New Jersey: Prentice-Hall.

No context found.

K. Castleman, Digital Image Processing. Prentice-Hall, 1997.

No context found.

K. R. Castleman. Digital Image Processing. Prentice-Hall, Inc., 1996.

No context found.

Castleman, K.R.: Digital Image Processing. Prentice Hall (1996) ISBN 0-13211467 -4.

No context found.

K. R. Castleman, Digital Image Processing. Englewood Cliffs, NJ: Prentice-Hall, 1996.

No context found.

Castleman, Kenneth R. DigitalImage Processing. Prentice Hall, 1979.

No context found.

K. R. Castleman. Digital Image Processing. Prentice Hall, 1996.

No context found.

Castleman, K. R., Digital Image Processing (Prentice Hall, New Jersey, 1996).

No context found.

Castleman, K., Digital Image Processing. Prentice Hall, New Jersey, 1996.

No context found.

Castleman, C.R.: Digital image processing, Prentice Hall, Upper Saddle River, New Jersey,

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC