Cascaded linear shift-invariant processors in optical pattern recognition

We study a cascade of linear shift-invariant processing modules (correlators), each augmented with a nonlinear threshold as a means to increase the performance of high-speed optical pattern recognition. This configuration is a special class of multilayer, feed-forward neural networks and has been proposed in the literature as a relatively fast best-guess classifier. However, it seems that, although cascaded correlation has been proposed in a number of specific pattern recognition problems, the importance of the configuration has been largely overlooked. We prove that the cascaded architecture is the exact structure that must be adopted if a multilayer feed-forward neural network is trained to produce a shift-invariant output. In contrast with more generalized multilayer networks, the approach is easily implemented in practice with optical techniques and is therefore ideally suited to the high-speed analysis of large images. We have trained a digital model of the system using a modified backpropagation algorithm with optimization using simulated annealing techniques. The resulting cascade has been applied to a defect recognition problem in the canning industry as a benchmark for comparison against a standard linear correlation filter, the minimum average correlation energy (MACE) filter. We show that the nonlinear performance of the cascade is a significant improvement over that of the linear MACE filter in this case.     

Face recognition by using feature-specific imaging

We present a face-recognition system based on the optical measurement of linear features. We describe a polarization-based optical system that computes linear projections of an incident irradiance distribution. We quantify the fundamental limitations of optical feature measurement. We find that higher feature fidelity can be obtained by feature-specific imaging than by postprocessing a conventional image. We present feature-fidelity results for wavelet, principal component, and Fisher features. We study face recognition by using a k-nearest neighbors classifier and two different feed-forward neural networks. Each image block is reduced to either a one- or a two-dimensional feature space for input to these recognition algorithms. As high as 99% recognition has been achieved with one-dimensional wavelet feature projections and 100% has been achieved with two-dimensional projections. A 95-fold increase in noise tolerance by use of feature-specific imaging has been demonstrated for an example of the face-recognition problem. An optical experiment is performed to validate these results.     

Classification of Fourier transform infrared microscopic imaging data of human breast cells by cluster analysis and artificial neural networks

Cluster analysis and artificial neural networks (ANNs) are applied to the automated assessment of disease state in Fourier transform infrared microscopic imaging measurements of normal and carcinomatous immortalized human breast cell lines. K-means clustering is used to implement an automated algorithm for the assignment of pixels in the image to cell and non-cell categories. Cell pixels are subsequently classified into carcinoma and normal categories through the use of a feed-forward ANN computed with the Broyden-Fletcher-Goldfarb-Shanno training algorithm. Inputs to the ANN consist of principal component scores computed from Fourier filtered absorbance data. A grid search optimization procedure is used to identify the optimal network architecture and filter frequency response. Data from three images corresponding to normal cells, carcinoma cells, and a mixture of normal and carcinoma cells are used to build and test the classification methodology. A successful classifier is developed through this work, although differences in the spectral backgrounds between the three images are observed to complicate the classification problem. The robustness of the final classifier is improved through the use of a rejection threshold procedure to prevent classification of outlying pixels.     

Food recognition using neural network classifier and multiple hypotheses image segmentation

ABSTRACT

This paper proposes the multiple-hypotheses image segmentation and feed-forward neural network classifier for food recognition to improve the performance. Initially, the food or meal image is given as input. Then, the segmentation is applied to identify the regions, where a particular food item is located using salient region detection, multi-scale segmentation, and fast rejection. Then, the features of every food item are extracted by the global feature and local feature extraction. After the features are obtained, the classification is performed for each segmented region using a feed-forward neural network model. Finally, the calorie value is computed with the aid of (i) food volume and (ii) calorie and nutrition measure based on mass value. The experimental results and performance evaluation are validated. The outcome of the proposed method attains 0.947 for Macro Average Accuracy (MAA) and 0.959 for Standard Accuracy (SA), which provides better classification performance.     

图片卫士:一个自动成人图像识别系统

设计并实现了一个自动识别成人图像识别系统“图片卫士”。图片卫士采用3层识别框架,利用肤色、纹理、图像视觉特征分层逐级识别成人图像。为了可靠地检测到图像中的肤色区域,提出了一种新的自适应统计肤色模型。在肤色检测基础上,通过皮肤纹理验证过程,图像中的人体皮肤区域被准确地分割出来。基于图像中皮肤区域,提取9个经验特征来表示图像内容,并采用AdaBoost算法构造一个总体分类器进行图像分类,识别正常图像和成人图像。在算法评估中,建立了一个78205幅图像的测试集,其中59885幅为正常图像,18320幅为成人图像。图片卫士显示了良好的系统性能,具有成人图像88．5％的识别率,正常图像92．5％的识别率。在PentiumⅣ1．5GHz的个人计算机上,图片卫士的平均处理速度为正常图像每秒5．6幅和成人图像每秒1．9幅。图片卫士可以应用在个人计算机或网络传输中,实时监控和过滤成人图像,还可以为网络安全等应用提供技术支持。     

Face recognition using Krawtchouk moment

Feature extraction is one of the important tasks in face recognition. Moments are widely used feature extractor due to their superior discriminatory power and geometrical invariance. Moments generally capture the global features of the image. This paper proposes Krawtchouk moment for feature extraction in face recognition system, which has the ability to extract local features from any region of interest. Krawtchouk moment is used to extract both local features and global features of the face. The extracted features are fused using summed normalized distance strategy. Nearest neighbour classifier is employed to classify the faces. The proposed method is tested using ORL and Yale databases. Experimental results show that the proposed method is able to recognize images correctly, even if the images are corrupted with noise and possess change in facial expression and tilt.     

Neural network that incorporates direct optical imaging

A compact neural network architecture is described that can be trained to sense and classify an optical image directly projected onto it. The system is based on the combination of a two-dimensional amorphous silicon photoconductor array and a liquid-crystal spatial light modulator. Appropriate filtering of the incident optical image on capture is incorporated into the network training rules through a modification of the standard backpropagation training algorithm. Training of the network on two image-classification problems is described: the recognition of handprinted digits and facial recognition. The network, once trained, is capable of stand-alone operation, sensing an incident image, and outputting a final classification signal in real time.     

Neural network classifier and multiple hypothesis image segmentation for dietary assessment using calorie calculator

Nowadays, dietary assessment becomes the emerging system for evaluating the person’s food intake. In this paper, the multiple hypothesis image segmentation and feed-forward neural network classifier are proposed for dietary assessment to enhance the performance. Initially, the segmentation is applied to input image which is used to determine the regions where a particular food item is located using salient region detection, multi-scale segmentation, and fast rejection. Then, the significant feature of food items is extracted by the global feature and local feature extraction method. After the features are obtained, the classification is performed for each segmented region using feed-forward neural network model. Finally, the calorie value is computed with the aid of (i) food area volume and (ii) calorie and nutrition measure based on mass value. The outcome of the proposed method attains 96% of accuracy value which provides the better classification performance.     

Optical implementation of neural networks for face recognition by the use of nonlinear joint transform correlators

We describe a nonlinear joint transform correlator-based two-layer neural network that uses a supervised learning algorithm for real-time face recognition. The system is trained with a sequence of facial images and is able to classify an input face image in real time. Computer simulations and optical experimental results are presented. The processor can be manufactured into a compact low-cost optoelectronic system. The use of the nonlinear joint transform correlator provides good noise robustness and good image discrimination.     

Automated detection of glioblastoma tumor in brain magnetic imaging using ANFIS classifier

This article proposes a novel and efficient methodology for the detection of Glioblastoma tumor in brain MRI images. The proposed method consists of the following stages as preprocessing, Non‐subsampled Contourlet transform (NSCT), feature extraction and Adaptive neuro fuzzy inference system classification. Euclidean direction algorithm is used to remove the impulse noise from the brain image during image acquisition process. NSCT decomposes the denoised brain image into approximation bands and high frequency bands. The features mean, standard deviation and energy are computed for the extracted coefficients and given to the input of the classifier. The classifier classifies the brain MRI image into normal or Glioblastoma tumor image based on the feature set. The proposed system achieves 99.8% sensitivity, 99.7% specificity, and 99.8% accuracy with respect to the ground truth images available in the dataset.     

Efficient impulse noise removal using hybrid neuro-fuzzy filter with optimized intelligent water drop technique

Impulse noise (IN) affects the digital image, during transmission, digital storage, and image acquisition. IN removal from an image is necessary as it retains the quality of the image. This work concentrates on the IN. A neuro-fuzzy (NF) system based on a fuzzy technique which is trained by a learning algorithm derived from neural network theory was implemented for the removal of noise. A NF network for noise filtering in grayscale images that combines two NF filters with a postprocessor to produce the output was presented. However, Sugeno-type is not intuitive technique and it also less accurate. To overcome these problems, a hybrid NF filter with optimized intelligent water drop (IWD) technique is introduced, where hybridized Sugeno–Mamdani-based fuzzy interference system is implemented in both the NF filters to obtain more efficient noise removal system. To improve the accuracy of the assignment of membership values to each input pixels, the optimized IWD technique is utilized, as the choice of membership function decides the efficiency of the noise removal in the images. Here, Fuzzy rules have been used to obtain the filtered output. The Hybrid method maintains the accuracy of the Sugeno model and also the interpretable capability of the Mamdani model. This method is robust against the IN and it is flexible, efficient, and accurate than existing filtering method in both noise attenuation and detail preservation and it has a great scope for better real-time applications.     

Photon-limited image classification with a feedforward neural network

An application of neural networks to the classification of photon-limited images is reported. A three-level feedforward network architecture is employed in which the input units of the network correspond to the pixels of a two-dimensional image. The network is trained in a minicomputer by the use of the backpropagation technique. The statistics of the network components are analyzed, resulting in a method by which the probability of correct classification of a given input image can be calculated. Photon-limited images of printed characters are obtained with a photon-counting camera and are classified. The experimental results are in excellent agreement with theoretical predictions.     

An action strategy generation framework for an on-line scheduling and control system in batch processes with neural networks

An on-line scheduling and control system in batch process management consists of three modules: a variability check module, action strategy generation module (ASGM) and corrective action module. ASGM is the key kernel of the above system, in which an appropriate modification mode is selected from alternative ones based on the plant status. In the proposed ASGM framework, a backpropagation neural network as a decision making sub-module is adopted, the preprocessor consisting of data collector, data filter, and data scale and the postprocessor as a simple distance-based classifier are developed to lead to significant improvement in recognition performance and detection of the 'unknown' class. The effectiveness of the proposed framework is demonstrated by experiments on two multipurpose batch plant case studies.     

支持向量机集成和特征选择联合算法

提出了两种基于支持向量机集成和特征选择联合算法。联合算法的核心思想是在构建基础分类器的同时选择有效特征。通过对实测舰船数据和公共数据的识别实验,证明了两种算法都可以用于舰船目标识别。算法一更适用于冗余特征较多的情况。算法二在对舰船目标识别时,选择的特征数目降低为原来特征数目的30%,正确分类率比单个支持向量机高近10%。     

Training of a neural network for image superresolution based on a nonlinear interpolative vector quantizer

Superresolution is the process of extending the spectrum of a diffraction-limited image beyond the optical passband. We consider the neural-network approach to accomplish superresolution and present results on simulated gray-scale images degraded by diffraction blur and additive noise. Images are assumed to be sampled at the Nyquist rate, which requires spatial interpolation for avoiding aliasing, in addition to frequency-domain extrapolation. A novel, to our knowledge, use of vector quantization for the generation of training data sets is also presented. This is accomplished by training of a nonlinear vector quantizer, whose codebooks are subsequently used in the supervised training of the neural network with backpropagation.     

Chained Dual-Generative Adversarial Network: A Generalized Defense Against Adversarial Attacks

Neural networks play a significant role in the field of image classification. When an input image is modified by adversarial attacks, the changes are imperceptible to the human eye, but it still leads to misclassification of the images. Researchers have demonstrated these attacks to make production self-driving cars misclassify Stop Road signs as 45 Miles Per Hour (MPH) road signs and a turtle being misclassified as AK47. Three primary types of defense approaches exist which can safeguard against such attacks i.e., Gradient Masking, Robust Optimization, and Adversarial Example Detection. Very few approaches use Generative Adversarial Networks (GAN) for Defense against Adversarial Attacks. In this paper, we create a new approach to defend against adversarial attacks, dubbed Chained Dual-Generative Adversarial Network (CD-GAN) that tackles the defense against adversarial attacks by minimizing the perturbations of the adversarial image using iterative oversampling and undersampling using GANs. CD-GAN is created using two GANs, i.e., CDGAN’s Sub-Resolution GAN and CDGAN’s Super-Resolution GAN. The first is CDGAN’s Sub-Resolution GAN which takes the original resolution input image and oversamples it to generate a lower resolution neutralized image. The second is CDGAN’s Super-Resolution GAN which takes the output of the CDGAN’s Sub-Resolution and undersamples, it to generate the higher resolution image which removes any remaining perturbations. Chained Dual GAN is formed by chaining these two GANs together. Both of these GANs are trained independently. CDGAN’s Sub-Resolution GAN is trained using higher resolution adversarial images as inputs and lower resolution neutralized images as output image examples. Hence, this GAN downscales the image while removing adversarial attack noise. CDGAN’s Super-Resolution GAN is trained using lower resolution adversarial images as inputs and higher resolution neutralized images as output images. Because of this, it acts as an Upscaling GAN while removing the adversarial attak noise. Furthermore, CD-GAN has a modular design such that it can be pre-fixed to any existing classifier without any retraining or extra effort, and can defend any classifier model against adversarial attack. In this way, it is a Generalized Defense against adversarial attacks, capable of defending any classifier model against any attacks. This enables the user to directly integrate CD-GAN with an existing production deployed classifier smoothly. CD-GAN iteratively removes the adversarial noise using a multi-step approach in a modular approach. It performs comparably to the state of the arts with mean accuracy of 33.67 while using minimal compute resources in training.     

An optical and electrical hybrid system for detecting abnormal masses on mammograms using Gabor feature extractor and neural network

Abstract

An optical and electrical hybrid recognition system for detecting abnormal masses on a mammogram image using a Gabor feature extractor and a neural network is presented. This prototype system detected all abnormal masses on 20 tested mammogram images without missing any.     

Neural network digital fringe calibration technique for structured light profilometers

We present a novel neural network signal calibration technique to improve the performance of triangulation-based structured light profilometers based on digital projection. The performance of such profilometers is often hindered by the capture of aberrated pattern intensity distributions, and hence we address this problem by employing neural networks in a signal mapping approach. We exploit the generalization and interpolation capabilities of a feed-forward backpropagation neural network to map from distorted fringe data to nondistorted data. The performance of the calibration technique is gauged both through simulation and experimentation, with simulation results indicating that accuracy can be improved by more than 80%. The technique requires just one image cross section for calibration and hence is ideal for rapid profiling applications.     

Evaluation of image quality of lumbar spine images: a comparison between FFE and VGA

PURPOSE: The aim of the present study is to compare two different methods for evaluation of the quality of clinical X-ray images. METHODS: Based on fifteen lumbar spine radiographs, two new sets of images were created. A hybrid image set was created by adding two distributions of artificial lesions to each original image. The image quality parameters spatial resolution and noise were manipulated and a total of 210 hybrid images were created. A set of 105 disease-free images was created by applying the same combinations of spatial resolution and noise to the original images. The hybrid images were evaluated with the free-response forced error experiment (FFE) and the normal images with visual grading analysis (VGA) by nine experienced radiologists. RESULTS: In the VGA study, images with low noise were preferred over images with higher noise levels. The alteration of the MTF had a limited influence on the VGA score. For the FFE study, the visibility of the lesions was independent of the sharpness and the noise level. No correlation was found between the two image quality measures. CONCLUSIONS: FFE is a precise method for evaluation of image quality, but the results are only valid for the type of lesion used in the study, whereas VGA is a more general method for clinical image quality assessment. The results of the FFE study indicate that there might be a potential to lower the dose levels in lumbar spine radiography without losing important diagnostic information.     

Invariant representation and hierarchical network for inspection of nuts from X-ray images

An X-ray based system for the inspection of pistachio nuts and wheat kernels for internal insect infestation is presented. The novelty of this system is twofold. First, we construct an invariant representation of infested nuts from X-ray images that is rich, robust, and compact. Insect infestation creates a tunnel, in the X-ray image, with reduced density of the natural material. The tunneling effect is encoded by linking troughs on the image and constructing a joint curvature-proximity distribution table for each nut. The latter step is designed to accentuate separation of those tunneling effects that are due to the natural structure of the nut. Second, since the representation is sparse, we partition the joint distribution table into several regions, where each region is used independently to train a backpropagation (BP) network. The outputs of these subnets are then collectively trained with another BP network. We show that the resulting hierarchical network has the advantage of reduced dimensionality while maintaining a performance similar to the standard BP network. © 1996 John Wiley & Sons, Inc.