B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Transactions on Neural Networks, 4(1):96--107, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....constituents of a wavelet network are single wavelets and their associated coefficients. We will consider the odd Gabor function as mother wavelet. It is well known that Gabor filters are recognized as good feature detectors and provide the best trade off between spatial and frequency resolution [11]. Considering the 2D image case, each single odd Gabor wavelet can be expressed as follows: sin ) 2 1 exp ) 0 1 x S x S x S x n i i i i T i i i y (1) where x represents image coordinates and n i = s , s , q, are parameters ....

B. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks, 4(1):96--107, 1993.

....pt go: o X ) exp cos(w0(x y) exp sin (w0(x y) 2.18) 2.19) In Fig. 2.3 plots of a 1 D and a 2 D odd Gabor function are shown. Gabor functions offer the best localization in both frequency and image space, and they are known to be good feature detectors [du Buf, 1993; Manjunath and Chellappa, 1993; Mehrotra et al. 1992; Michaelis, 1997] In this thesis we will use odd Gabor functions only, as they have proven to give the best results for the purposes we will use them for. We will discuss this topic in more detail in Section 3.1. 2.2 Introduction to Gabor Wavelet Networks In this section ....

....[Daugman, 1985; Jones and Palmer, 1987] In fact, it has been suggested [Daugman, 1988; Porat and Zeevi, 1988] that the receptive field responses of simple cells can be described by the family of 2 D Gabor wavelets. In addition, Gabor filters are recognized as good feature detectors [du Buf, 1993; Manjunath and Chellappa, 1993; Mehrotra et al. 1992; Michaelis, 1997] Especially for ao:0 2, they are often used for edge detection [Michaelis, 1997] Specific uses of the odd Gabor function have particular advantages, which will be discussed in Chapter 3. An image representation using GWNs has the advantage of being ....

[Article contains additional citation context not shown here]

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks', 4(1):96-107, 1993.

....as orientation, position and scale) are individually optimized to reflect the particular local image structure. Gabor Wavelet Networks have several advantages: 1) GWN allow an efficient and sparse coding while coding is adaptive to the task at hand. 2) Gabor filters are good feature detectors [6] and the optimized parameters of each of the Gabor wavelets are directly related to the underlying image structure. 3) The wavelet coefficients (or weights) of each of the Gabor wavelets are linearly related to the filter responses and with that they are also directly related to the underlying ....

....results depend on the number of filters used. The last Section concludes with some final remarks. 2 Introduction to Gabor Wavelet Networks Wavelet Networks were first introduced by [10] and the use of Gabor functions is inspired by the fact that they are recognized to be good feature detectors [6]. To define a GWN, we start by taking a set of N odd, real valued Gabor wavelet functions # = # n 1 , # of the form # n x, y = exp # s y ( x c x ) sin # (y c y ) cos #) sin , 1) with n = c x , c y , #, s x , s y ) Here, c x , ....

[Article contains additional citation context not shown here]

B. Manjunath, R. Chellappa, A unified approach to boundary perception: edges, textures, and illusory contours, IEEE Trans. Neural Networks 4 (1) (1993) 96--107.

....combination of Gabor wavelets where the parameters of each of the Gabor functions (such as orientation, position and scale) are optimized to reflect the particular local image structure. The use of Gabor Wavelet Networks has several advantages, namely: 1. Gabor filters are good feature detectors [8, 9] and the optimized parameters of each Gabor wavelet reflect the underlying image structure, 2. The Gabor wavelet weights are directly related to the Gabor filter responses and thus also reflect the underlying local image structure, 3. The precision of the representation can be varied to any ....

....and conclude the paper with the experiments in section 4 and concluding remarks. 2. Introduction to Gabor Wavelet Networks The basic idea of the wavelet networks was first stated in [11] and the use of Gabor functions is inspired by the fact that they are recognized to be good feature detectors [8, 9]. Number of Wavelets 116 216 original 16 52 Figure 1. Face images reconstructed with different number of wavelets To define a GWN, we start out, generally speaking, by taking a family of N odd Gabor wavelet functions # = of the form #n x, y exp s y ( x c x ) sin ....

B. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks, 4(1):96--107, 1993.

....representation. The idea of the wavelet network is inspired by [15] and the use of Gabor functions is inspired by the fact that they provide the best possible trade off between spatial resolution and frequency resolution. Furthermore, the Gabor filters are recognized to be good feature detectors [9]. An image representation with Gabor Wavelet Networks has the advantage of being sparser than the Gabor jet representation [13] Yet, it allows to encodes almost the entire image information and allows a good reconstruction. To define a GWN, we start out, generally speaking, by taking a family of ....

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks, 4(1):96-107, 1993.

....combination of Gabor wavelets where the parameters of each of the Gabor functions (such as orientation, position and scale) are optimized to reflect the particular local image structure. The use of Gabor Wavelet Networks has several advantages, namely: 1. Gabor filters are good feature detectors [8, 9] and the optimized parameters of each Gabor wavelet reflect the underlying image structure, 2. The Gabor wavelet weights are directly related to the Gabor filter responses and thus also reflect the underlying local image structure, 3. The precision of the representation can be varied to any ....

....and conclude the paper with the experiments in section 4 and concluding remarks. 2. Introduction to Gabor Wavelet Networks The basic idea of the wavelet networks was first stated in [11] and the use of Gabor functions is inspired by the fact that they are recognized to be good feature detectors [8, 9]. Number of Wavelets 116 216 original 16 52 Figure 1. Face images reconstructed with di erent number of wavelets To define a GWN, we start out, generally speaking, by taking a family of N odd Gabor wavelet functions = f n1 ; nN g of the form n x; y = exp 1 2 h s x ( x ....

B. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks, 4(1):96--107, 1993.

....into a connected region if all the pixels in the region satisfy some conditions. One of the apparent deficits with these algorithms is their iterative (serial) nature (Liou et al. 1991) There are some recent algorithms which are partially parallel (Liou et al. 1991; Mohan and Nevatia 1992; and Manjunath and Chellappa 1993). Most of these techniques rely on domain specific heuristics to perform segmentation, and no unified computational framework exists to explain the general phenomenon of scene segmentation (Haralick and Shapiro 1985) The problem of scene segmentation is computationally hard (Gurari and Wechsler ....

Manjunath, B. S., and Chellappa, R. 1993. A unified approach to boundary perception: Edges, textures, and illusory contours. IEEE Trans. Neural Net. 4, 96-108.

....are optimized to reflect the particular local image structure. Gabor wavelet networks have several advantages: 1. By their very nature, Gabor wavelet networks are invariant to some degree to affine deformations and homogeneous illumination changes, 2. Gabor filters are good feature detectors [13] and the optimized parameters of each of the Gabor wavelets are directly related to the underlying image structure, 3. the weights of each of the Gabor wavelet are directly related to their filter responses and with that they are also directly related to the underlying local image structure, 4. ....

.... the image I , represented with 16, 52, 116 and 216 Gabor wavelets (left to right) 2 Introduction to Gabor Wavelet Networks The basic idea of the wavelet networks is first stated by [27] and the use of Gabor functions is inspired by the fact that they are recognized to be good feature detectors [13]. To define a GWN, we start out, generally speaking, by taking a family of N odd Gabor wavelet functions = f n1 ; nN g of the form n x; y = exp 1 2 h sx ( x cx) cos (y cy ) sin ) i 2 h sy ( x cx) sin (y cy ) cos ) i 2 sin sx ( x cx) cos (y ....

[Article contains additional citation context not shown here]

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks, 4(1):96--107, 1993.

.... a small set of such filters, differently oriented, can be used to evidence texture differences in an image by using the responses of networks selective to various orientations (i.e. lying along different axes) Turner1986] Malik and Perona1990] Bergen and Landy1991] Jain and Farrokhnia1991] Manjunath and Chellappa1993] Indiveri et al..1995] the presence of repeated oriented elements shows up as the strongest output for the networks of the corresponding orientations. Stereopsis Stereoscopic depth estimation is mainly based on disparity measurement. Disparity is a local property that can be associated to a ....

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: Edges, textures and illusory contours. IEEE Trans. Neural Net., 4:96--107, 1993.

....list of tasks when the simple considerd here is, at least roughly applicable, can be extended. In particular, the segmentation of the images and analysis of the remotely sensed scenes fall into this cathegory. We hope that analysis of image processing tasks contained in , e.g. 3] 1] 6] and [5] will be a reach source of examples of possible applications for the proposed model. We remark that statistical pattern recognition task, as stated e.g. in [2] is essentially different than that considred here. The rest of the paper is organized as follows. Assumptions concerning the model are ....

B. S. Manjunath and R. Chellapa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks, vol. 4:96--107, 1993.

....methods trying to imitate human performance. Studies of the human visual system support the conjecture that this segmentation process is purely pre attentive [18, 2, 28] occurring at early visual stages [30, 16, 7, 31] We have extended existing schemes for boundary perception and completion [29, 8, 9, 27, 19, 10, 21, 26, 32] by a relaxation labeling mechanism [3, 4] which groups the phases of parametric neural oscillators into perceptual objects exploiting simple geometric constraints between the basic features. The proposed model consists of three successive processing stages which model some of the mechanisms ....

B. S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures and illusory contours. IEEE Transactions on Neural Networks, 4(1):96--107, 1993.

....text ( Pi) and nontext ( in the subspace after LDA projection. Text and nontext overlap. choice as a classifier because of its ability to learn. Theoretically, a three layer neural network can approximate any nonlinear function after training. The success of neural networks in related problems [8, 11, 13, 39, 48] provides us with further motivation to rely on a neural network as a classifier to identify text regions. Our methodology uses a small window (typically 16 Theta 16) to scan the image and classify each window as text or non text using the neural network (Figure 2) We will address the following ....

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures and illusory contours. IEEE Trans. Neural Networks, 4:96 -- 108, 1993.

No context found.

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Transactions on Neural Networks, 4(1):96--107, 1993.

No context found.

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Transactions on Neural Networks, 4(1):96--107, 1993.

No context found.

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Transactions on Neural Networks, 4(1):96--107, 1993.

No context found.

B. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks, 4(1):96--107, 1993.

No context found.

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. Neural Networks, 4(1):96--107, 1993.

No context found.

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Transactions on Neural Networks, 4(1):96--107, 1993.

No context found.

B.S. Manjunath and R. Chellappa. A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Transactions on Neural Networks, 4(1):96--107, 1993.

No context found.

Manjunath DS, Chellappa R, A unified approach to boundary perception:edges, textures and illusory contours, IEEE Trans. on Neural Networks, vol 4, pp. 96-107 (1993) 5

No context found.

Manjunath BS, Chellappa R (1993) A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans Neural Networks 4:96--107

No context found.

Manjunath, B.S., and Chellappa, R. (1993). A unified approach to boundary perception: edges, textures, and illusory contours. IEEE Trans. on Neural Networks, 4, 96-108.

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