An MLP-based texture segmentation method without selecting a feature set

doi:10.1016/S0262-8856(97)00035-8

Image and Vision Computing

Volume 15, Issue 12, December 1997, Pages 937-948

https://doi.org/10.1016/S0262-8856(97)00035-8 Get rights and content

Abstract

A texture segmentation technique which employs a multilayer perceptron (MLP) and does not consider the selection of features is presented in this paper. Thus, users can avoid selection and computation of the feature set and hence real-time segmentation may be possible. The technique apparently works in a fashion similar to our visual system whereby we do not consciously compute any feature for texture discrimination. A detailed study has been made for the selection of the network size. A newly proposed variant of the back-propagation algorithm has been used for more efficient training of the network. An edge-preserving noise-smoothing approach has been proposed to remove noise from the segmented image.

References (25)

T.R. Reed et al.
A review of recent texture segmentation and feature extraction techniques
Computer Vision, Graphics and Image Processing: Image Understanding
(1993)
P.C. Chen et al.
Segmentation by texture using co-occurrence matrix and split-and-merge algorithm
Computer Graphics and Image Processing
(1979)
J. Mao et al.
Texture classification and segmentation using simultaneous autoregressive model
Patt. Recog.
(1992)
C. Shang et al.
Principal feature-based texture classification with neural networks
Patt. Recog.
(1994)
U. Bhattacharya et al.
On the rate of convergence of perceptron learning
Patt. Recog. Lett.
(1995)
R.A. Jacobs
Increased rates of convergence through learning rate adaptation
Neural Networks
(1988)
K. Hornik et al.
Multilayer feedforward networks are universal approximators
Neural Networks
(1989)
M. Nagao et al.
Edge preserving smoothing
Computer Graphics and Image Processing
(1979)
R.M. Haralick
Statistical and structural approaches to texture
L.v. Gool et al.
Texture analysis anno 1983
Computer Vision, Graphics and Image Processing
(1985)

M. Tuceryan et al.

Texture analysis

R.M. Haralick et al.

Textural features for image classification

IEEE Trans. Syst. Man Cybernetics

(1973)

Cited by (11)

Automatic brain segmentation using artificial neural networks with shape context
2018, Pattern Recognition Letters
Segmenting brain tissue from MR scans is thought to be highly beneficial for brain abnormality diagnosis, prognosis monitoring, and treatment evaluation. Many automatic or semi-automatic methods have been proposed in the literature in order to reduce the requirement of user intervention, but the level of accuracy in most cases is still inferior to that of manual segmentation. We propose a new brain segmentation method that integrates volumetric shape models into a supervised artificial neural network (ANN) framework. This is done by running a preliminary level-set based statistical shape fitting process guided by the image intensity and then passing the signed distance maps of several key structures to the ANN as feature channels, in addition to the conventional spatial-based and intensity-based image features. The so-called shape context information is expected to help the ANN to learn local adaptive classification rules instead of applying universal rules directly on the local appearance features. The proposed method was tested on a public datasets available within the open MICCAI grand challenge (MRBrainS13). The obtained average Dice coefficient were 84.78%, 88.47%, 82.76%, 95.37% and 97.73% for gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), brain (WM + GM) and intracranial volume respectively. Compared with other methods tested on the same dataset, the proposed method achieved competitive results with comparatively shorter training time.
Multi-level pixel-based texture classification through efficient prototype selection via normalized cut
2010, Pattern Recognition
Citation Excerpt :
Pixel-based texture classification [3] is a combination of the two aforementioned application domains in the sense that it determines the class to which every pixel of an input image belongs, which obviously leads to the segmentation of the image as a collateral effect. Hence, pixel-based classification is a possible way to perform supervised segmentation [4,19–25]. In pixel-based texture classification, several measures are computed for each image pixel by applying a set of texture feature extraction methods to its neighboring pixels.
This paper presents a new efficient technique for supervised pixel-based classification of textured images. A prototype selection algorithm that relies on the normalized cut criterion is utilized for automatically determining a subset of prototypes in order to characterize each texture class at the local level based on the outcome of a multichannel Gabor filter bank. Then, a simple minimum distance classifier fed with the previously determined prototypes is used to classify every image pixel into one of the given texture classes. Multi-sized evaluation windows following a top-down approach are used during classification in order to improve accuracy near frontiers of regions of different texture. Results with standard Brodatz, VisTex and MeasTex compositions and with complex real images are presented and discussed. The proposed technique is also compared with alternative texture classifiers.
Integrating region and edge information for texture segmentation using a modified constraint satisfaction neural network
2008, Image and Vision Computing
Citation Excerpt :
The algorithm requires the probability values for all the pixels corresponding to each class in an image and edge maps at two different thresholds as input. The initial class probabilities can be obtained using any texture segmentation algorithm such as [4,29–31,13,32,14,33]. The initial edge maps can be obtained using texture edge detection techniques such as [15,18,34,35].
In this paper, we propose an approach for texture segmentation by integrating region and edge information. The algorithm uses a constraint satisfaction neural network for texture segmentation with additional edge constraints. Initial class probabilities and edge maps are computed using multi-channel, multi-resolution filters to obtain image segmented map and edge map. The complementary information of the segmented map and the edge map are iteratively updated using a modified CSNN to satisfy a set of constraints to obtain superior segmentation results.The proposed methodology is tested on simulated as well as natural textures and it produces satisfactory results. The proposed methodology is also tested on a synthetic aperture radar (SAR) image.
Multiresolution genetic clustering algorithm for texture segmentation
2003, Image and Vision Computing
Citation Excerpt :
Texture segmentation is usually cast as an optimization problem. A wide variety of texture segmentation techniques have been reported in the literature [1,2,4–6,12,14,15,18,21,23,26]. Among those techniques, Bayesian approaches based on Markov random field (MRF) is one of the most frequently used [1,5,6,7,12,18,26].
This work plans to approach the texture segmentation problem by incorporating genetic algorithm and K-means clustering method within a multiresolution structure. As the algorithm descends the multiresolution structure, the coarse segmentation results are propagated down to the lower levels so as to reduce the inherent class–position uncertainty and to improve the segmentation accuracy. The procedure is described as follows. In the first step, a quad-tree structure of multiple resolutions is constructed. Sampling windows of different sizes are utilized to partition the underlying image into blocks at different resolution levels and texture features are extracted from each block. Based on the texture features, a hybrid genetic algorithm is employed to perform the segmentation. While the select and mutate operators of the traditional genetic algorithm are adopted in this work, the crossover operator is replaced with K-means clustering method. In the final step, the boundaries and the segmentation result of the current resolution level are propagated down to the next level to act as contextual constraints and the initial configuration of the next level, respectively.
Multiresolution image segmentation integrating Gibbs sampler and region merging algorithm
2003, Signal Processing
Citation Excerpt :
This requirement is sensible because if the training phase were needed, than when textures outside the training set present in the images, the algorithm would have difficulty identifying them. In the last few decades, a wide variety of image segmentation techniques have been reported in the literature [1,3,4,7,8,10–15,17]. Among them, Markov random field is one of the most frequently utilized techniques [1,4,7,8,12,13,17].
This work approaches the texture segmentation problem by incorporating Gibbs sampler (i.e., the combination of Markov random fields and simulated annealing) and a region-merging process within a multiresolution structure with “high class resolution and low boundary resolution” at high levels and “low class resolution and high boundary resolution” at lower ones. As the algorithm descends the multiresolution structure, the coarse segmentation results are propagated down to the lower levels so as to reduce the inherent class-boundary uncertainty and to improve the segmentation accuracy. The computational complexity and frequent occurrences of over-segmentation of Gibbs sampler are addressed and the computationally and functionally effective region-merging process is included to allow Gibbs sampler to start its annealing schedule at relatively low pseudo-temperature and to guide the search trajectory away from local minima associated with over-segmented configurations.
Neural network-based text location in color images
2001, Pattern Recognition Letters
This paper proposes neural network-based text locations in complex color images. Texture information extracted on several color bands using neural networks is combined and corresponding text location algorithms are then developed. Text extraction filters can be automatically constructed using neural networks. Comparisons with other text location methods are presented; indicating that the proposed system has a better accuracy.

View all citing articles on Scopus

View full text

An MLP-based texture segmentation method without selecting a feature set

Abstract

Computer Vision, Graphics and Image Processing: Image Understanding

Computer Graphics and Image Processing

Patt. Recog.

Patt. Recog.

Patt. Recog. Lett.

Neural Networks

Neural Networks

Computer Graphics and Image Processing

Statistical and structural approaches to texture

Texture analysis anno 1983

Computer Vision, Graphics and Image Processing

Texture analysis

Textural features for image classification

IEEE Trans. Syst. Man Cybernetics