Non-uniform Lighting Conditions Research Articles

Implementation of a High-Accuracy Neural Network-Based Pupil Detection System for Real-Time and Real-World Applications.

In this paper, the implementation of a new pupil detection system based on artificial intelligence techniques suitable for real-time and real-word applications is presented. The proposed AI-based pupil detection system uses a classifier implemented with slim-type neural networks, with its classes being defined according to the possible positions of the pupil within the eye image. In order to reduce the complexity of the neural network, a new parallel architecture is used in which two independent classifiers deliver the pupil center coordinates. The training, testing, and validation of the proposed system were performed using almost 40,000 eye images with a resolution of 320 × 240 pixels and coming from 20 different databases, with a high degree of generality. The experimental results show a detection rate of 96.29% at five pixels with a standard deviation of 3.38 pixels for all eye images from all databases and a processing speed of 100 frames/s. These results indicate both high accuracy and high processing speed, and they allow us to use the proposed solution for different real-time applications in variable and non-uniform lighting conditions, in fields such as assistive technology to communicate with neuromotor-disabled patients by using eye typing, in computer gaming, and in the automotive industry for increasing traffic safety by monitoring the driver's cognitive state.

Segmentation and deep learning to digitalize the kinematics of flow-type landslides

Flow-type landslides, including subaerial and submarine debris flows, have poor spatiotemporal predictability. Therefore, researchers rely heavily on experimental evidence in revealing complex flow mechanisms and evaluating theoretical models. To measure the velocity field of experimental flows, conventional image analysis tools for measuring soil deformation and hydraulics have been borrowed. However, these tools were not developed for capturing the kinematics of fast-moving soil–water mixtures over complex terrain under non-uniform lighting conditions. In this study, a new framework based on deep learning was used to automatically digitalize the kinematics of experimental flow-type landslides. Captured images were broken into sequences and binarized using a fully convolutional neural network (FCNN). The proposed framework was demonstrated to outperform classic image processing algorithms (e.g., particle image velocimetry, trainable Weka segmentation, and thresholding algorithms) over a wide range of experimental conditions. The FCNN model was even able to process images from consumer-grade cameras under complex shadow, light, and boundary conditions. This feature is most useful for field-scale experimentation. With fewer than 15 annotated training images, the FCNN digitalized experimental flows with an accuracy of 97% in semantic segmentation.

Hetero-associative Memory Based New Iraqi License Plate Recognition

As a result of recent developments in highway research as well as the increased use of vehicles, there has been a significant interest paid to the most current, effective, and precise Intelligent Transportation System (ITS). In the field of computer vision or digital image processing, the identification of specific objects in an image plays a crucial role in the creation of a comprehensive image. There is a challenge associated with Vehicle License Plate Recognition (VLPR) because of the variation in viewpoints, multiple formats, and non-uniform lighting conditions at the time of acquisition of the image, shape, and color, in addition, the difficulties like poor image resolution, blurry image, poor lighting, and low contrast, these must be overcome. This paper proposed a model by using Modify Bidirectional Associative Memory (MBAM), which is one type of Hetero-associative memory, MBAM works in two phases (learning and convergence phases) to recognize the number plate, and this proposed model can overcome these difficulties because MBAM's associative memory has a high ability to accept noise and distinguish distorted images, as well as the speed of the calculation process due to the small size of the network. The accuracy of plate region localization is 99.6%, the accuracy for character segmentation is 98%, and the achieved accuracy for character recognition is 100% in various circumstances

Multiple linear regression based illumination normalization for non-uniform light image thresholding

Thresholding-based two-class image binarization is one of the simplest and most popular approaches. However, the performance of global thresholding degrades under non-uniform lighting conditions. Local thresholding methods are widely used for binarizing uneven light images. The appropriate choice of the initial size of the window and designing the bimodal criteria function are the most challenging tasks for the local thresholding approaches. Therefore, to make it simpler, in this work, a novel approach is developed to improve the efficacy of binarizing any uneven light images. To begin with, a two stage approach is developed to extract valid training sample points from the uneven light images for estimating the illumination surface. In addition, the Multiple-Linear-Regression (MLR) method is applied on the extracted training sample points to estimate the illumination surface. Furthermore, the estimated illumination surface is used to normalize the non-uniform light of the image to binarize the image using Otsu’s global thresholding. The proposed approach is validated on different variants of uneven light images and with six different states of art uneven light image binarization approaches. It is observed from the simulations that the performance of the proposed approach outperforms the other approaches in qualitative as well as quantitative measures. Further, the binarization of uneven document image methods are not effective on object background binarization of uneven images. The proposed approach has the average F-Measure (F1) score of 0.98, average Jaccard Index (JI) score of 0.97, average Percentage of Misclassification Error (PME) score of 1.10 and the computational complexity of 2.64 sec.

Non-Uniform-Illumination Image Enhancement Algorithm Based on Retinex Theory

To address the issues of fuzzy scene details, reduced definition, and poor visibility in images captured under non-uniform lighting conditions, this paper presents an algorithm for effectively enhancing such images. Firstly, an adaptive color balance method is employed to address the color differences in low-light images, ensuring a more uniform color distribution and yielding a low-light image with improved color consistency. Subsequently, the image obtained is transformed from the RGB space to the HSV space, wherein the multi-scale Gaussian function is utilized in conjunction with the Retinex theory to accurately extract the lighting components and reflection components. To further enhance the image quality, the lighting components are categorized into high-light areas and low-light areas based on their pixel mean values. The low-light areas undergo improvement through an enhanced adaptive gamma correction algorithm, while the high-light areas are enhanced using the Weber–Fechner law for optimal results. Then, each block area of the image is weighted and fused, leading to its conversion back to the RGB space. And a multi-scale detail enhancement algorithm is utilized to further enhance image details. Through comprehensive experiments comparing various methods based on subjective visual perception and objective quality metrics, the algorithm proposed in this paper convincingly demonstrates its ability to effectively enhance the brightness of non-uniformly illuminated areas. Moreover, the algorithm successfully retains details in high-light regions while minimizing the impact of non-uniform illumination on the overall image quality.

Barcode positioning method based on line detection under non-uniform illumination

The positioning problem of the barcode is easily affected by factors such as uneven illumination and complex background. Based on the EDLines line detection algorithm, a positioning algorithm that can deal with the barcode positioning problem under non-uniform lighting conditions is proposed. First, the EDLines algorithm is used to detect the straight-line segment of the image, and the midpoint coordinates of all line segments are extracted. Since the midpoint coordinates near the barcode are relatively densely distributed, the density-based HDBSCAN point clustering algorithm is used for all midpoint coordinates for density clustering. Then a filtering strategy based on the consistency of line segment direction is proposed to preliminarily filter the clustering results, and one or more point clustering results that may belong to the barcode area are obtained. Then the RANSAC algorithm is used to perform line fitting on the point clustering results. The accuracy of the straight-line fitting results is judged by direction consistency screening, and the intersection of the fitted straight line and the straight-line segment to which the centroid belongs is calculated for the results that meet the expectations, and the width of the barcode area is obtained. Finally, the straight-line segment obtained by screening is calculated. The average length and the height of the barcode area are got, to accurately locate the barcode position. Compared with the previous algorithms on the public barcode dataset and the self-made barcode dataset, the algorithm has good positioning accuracy and running speed.

Lightness Perception of Fabrics Under Non-uniform and Uniform Lighting Conditions

Lightness Perception of Fabrics Under Non-uniform and Uniform Lighting Conditions

MIXED REALITY CONTENT ALIGNMENT IN MONUMENTAL ENVIRONMENTS

Abstract. Mixed reality provides on-the-spot and real-time data access capabilities by making virtual models and information more intuitive and accessible. Furthermore, allowing the operator to perceive 3D as holograms would allow for a more natural and straightforward manipulation of the perceived 3D content by permitting the augmentation of real objects with various levels of data. This can be accomplished by appropriately registering and superimposing the presented 3D models with the surrounding environment. This work aims to provide a quantitative evaluation of HoloLens 2 capabilities in registering virtual content inside monumental spaces. Two different methodologies are evaluated: Vuforia image targets and Microsoft World Locking Tools (WLTs). Tests have been performed inside Milan Cathedral's monumental spaces. Here, ambience dimensions, single architectural element repetition and non-uniform lighting conditions push out-of-the-box methods to their limits. Results show that WLTs with their space pins API can correctly reference virtual content keeping deviations in the order of 15 cm coping with the scale error produced from sensors’ drifts.

A new scheme of polar Fast Fourier Transform Code for iris recognition through symbolic modelling approach

The rich random structure of the iris, non-invasiveness and temporal stability properties makes it one of the most powerful biometric trait for person identification. Iris recognition techniques have demonstrated very low false acceptance rates and very high matching efficiency in large iris datasets. However, iris recognition systems face several challenges including non-uniform lighting conditions, occlusion of the eye by the eyelid, variation of head pose (causes dimensional inconsistencies of iris part), the reflection of surrounding objects and many others. The paper has addressed some of the reported issues of iris recognition technique by exploring new feature extraction technique and devising a new symbolic modelling approach for feature representation and classification, along with other classification techniques namely k-NN, SVM and PNN. The symbolic approach employed in the proposed research to represent iris feature reduces the dimensionality of feature space and also reduces the actual time required for iris recognition. The symbolic approach with new feature extraction technique has attained recognition accuracy of 100%, 98.26%, 99.25% and 96.67% on CASIA 1.1, CASIA 4.0 Interval, SGGSIE&T and the newly formed VISA Iris datasets respectively. Additionally, authors deployed previous algorithms and their findings of the results are discussed in comparison with proposed techniques.

Editorial

Editorial

A robust methodology for outdoor optical mark recognition

Outdoor optical mark recognition is an extremely useful tool for recognition of large industrial equipment and application of computer vision-based systems for tracking and positioning. However, current algorithms rely on thresholding and corner detection to identify checkerboard-like patterns, which is not appropriate for non-uniform lighting conditions. This paper presents a robust methodology to identify optical markers in outdoor environments. A GPU-based region filling algorithm automatically detects all contiguous color regions without computing seed points. Post-processing steps extract high-level information from these regions. Analysis of identified contiguous color region allows simultaneous identification of all checkerboard and targets (concentric regions) in the scene. Analysis of variance demonstrates that the proposed methodology is robust to lighting, environment, perspective, and occlusion. Tests indicate that precision and recall for checkerboard and target identification in outdoor conditions are expected to be above 97%. The parallel algorithm implementation using OpenCL yields better results and is two times faster than previous region filling algorithms, taking about 0.6 s to process a full-HD picture using modern hardware

A segmentation-based adaptive image enhancement method inspired by the self-adjust features of HVS

This paper focuses on the enhancement of the images captured in insufficient or non-uniform lighting conditions. Inspired by the self-adjust features of HVS (Human Vision System), a Segmentation-Based Adaptive Image Enhancement (SAIE) method is proposed to improve the visibility of degraded images. First the local characteristics of different parts of each image (called sub-block images) obtained by image segmentation are described by Gaussian curvature, then the parameters of Gaussian filter for each sub-block image are chosen adaptively according to its local characteristics to estimate the illumination component of the whole image accurately. After that, adaptive dynamic range adjustment for the estimated illumination based on global characteristics and contrast enhancement for sub-block images are implemented. Finally, the enhancement result is obtained after color adjustment. The experiments on five typical image sets show that SAIE can provide effective and robust enhancement results. Comparative qualitative and quantitative evaluations with MSRCR, IRME and MSRCR+AL are also provided to demonstrate the efficiency and superiority of the proposed method.

Unsupervised eye pupil localization through differential geometry and local self-similarity matching.

The automatic detection and tracking of human eyes and, in particular, the precise localization of their centers (pupils), is a widely debated topic in the international scientific community. In fact, the extracted information can be effectively used in a large number of applications ranging from advanced interfaces to biometrics and including also the estimation of the gaze direction, the control of human attention and the early screening of neurological pathologies. Independently of the application domain, the detection and tracking of the eye centers are, currently, performed mainly using invasive devices. Cheaper and more versatile systems have been only recently introduced: they make use of image processing techniques working on periocular patches which can be specifically acquired or preliminarily cropped from facial images. In the latter cases the involved algorithms must work even in cases of non-ideal acquiring conditions (e.g in presence of noise, low spatial resolution, non-uniform lighting conditions, etc.) and without user's awareness (thus with possible variations of the eye in scale, rotation and/or translation). Getting satisfying results in pupils' localization in such a challenging operating conditions is still an open scientific topic in Computer Vision. Actually, the most performing solutions in the literature are, unfortunately, based on supervised machine learning algorithms which require initial sessions to set the working parameters and to train the embedded learning models of the eye: this way, experienced operators have to work on the system each time it is moved from an operational context to another. It follows that the use of unsupervised approaches is more and more desirable but, unfortunately, their performances are not still satisfactory and more investigations are required. To this end, this paper proposes a new unsupervised approach to automatically detect the center of the eye: its algorithmic core is a representation of the eye's shape that is obtained through a differential analysis of image intensities and the subsequent combination with the local variability of the appearance represented by self-similarity coefficients. The experimental evidence of the effectiveness of the method was demonstrated on challenging databases containing facial images. Moreover, its capabilities to accurately detect the centers of the eyes were also favourably compared with those of the leading state-of-the-art methods.

Image enhancement based on contourlet transform

Images captured with insufficient illumination generally have dark shadows and low contrast. This problem seriously affects other forms of image processing schemes such as face detection, security surveillance, image fusion. In this paper, a new image enhancement algorithm using the important features of the contourlet transform is presented. A new transformation function is developed based on the existing sigmoid function and the tanh functions which have very interesting properties in enhancing images which are suffering from low illuminations or non-uniform lighting conditions. Literature dictates that contourlet transform has better performance in representing the image salient features such as edges, lines, curves, and contours than wavelets for its anisotropy and directionality and is therefore well suited for multiscale edge-based image enhancement. The algorithm works for gray scale and color images. For a color image, it is first converted from RGB (red, green, and blue) to HSI (hue, saturation, and intensity) color model. Then, the intensity component of the HSI color space is adjusted the preserving the original color using a new nonlinear transformation function. The simulation results show that this approach gives encouraging results for images taken in low-light and/or non-uniform lighting conditions. The results obtained are compared with other enhancement algorithms based on wavelet transform, curvelet transform, bandlet transform, histogram equalization (HE), and contrast limited adaptive histogram equalization. The performance of the enhancement based on the contourlet transform method is superior. The algorithm is checked for a total of 151 test images. A total of 120 of them are used for subjective evaluation and 31 are used for objective evaluation. For over 90 % of the cases, the system is superior over the other enhancement methods.

Improved Image Analysis Based System to Reliably Predict the Live Weight of Pigs on Farm: Preliminary Results

A computer vision system was developed to automatically measure the live weight of pigs without human intervention. The system was trialled on both research and commercial farms to demonstrate the ability of the system to cope with different conditions and non-uniform lighting conditions. Early results demonstrate that the system can achieve sufficient practical accuracy. The results of the initial trials demonstrated that weight of the pigs can be predicted with an average error of 1.18 kg. Precision, reliability and repeatability are likely to be increased in future through improved weight prediction models, increased image resolution and algorithm enhancement.

A modif ied image enhancement algorithm based on color constancy

According to the color constancy theory, a modified variation Retinex is proposed for improving the visibility of the dark regions in images under insufficient and/or non-uniform lighting conditions. A new penalty functional based on nonlinear diffusion and correlation between the reflectance and the given image is designed for the intensity image enhancement, followed by adaptive color compensation. With high computational efficiency achieved by an improved multi-resolution algorithm, simulation results prove that the proposed method shows more colorful and vivid visual performance, and achieves wider dynamic range with higher objective standard values.

A locally tuned nonlinear technique for color image enhancement

An innovative technique for the enhancement of digital color images captured under extremely non-uniform lighting conditions is proposed in this paper. The key contributions of this technique are adaptive intensity enhancement, contrast enhancement and color restoration. Simultaneous enhancement of extreme dark and bright intensity regions in an image is performed by a specifically designed Locally Tuned Sine Non-Linear (LTSN) function. The intensity of each pixel's magnitude is tuned based on its surrounding pixels to accomplish contrast enhancement. Retrieval of the color information from the enhanced intensity image is achieved by a linear color restoration process which is based on the chromatic information of the input image. Experimental evaluations show that the proposed algorithm can be effectively used for improving the visibility of night time surveillance video sequences with frames having extreme bright and dark regions.

A tunable high-performance architecture for enhancement of stream video captured under non-uniform lighting conditions

A novel architecture for performing hue-saturation-value (HSV) domain enhancement of digital color images captured under non-uniform lighting conditions is proposed in this paper for video streaming applications. The approach promotes log-domain computation to eliminate all multiplications, divisions and exponentiations utilizing the compact high-speed logarithmic estimation modules. An optimized quadrant symmetric architecture is incorporated into the design of homomorphic filter for the enhancement of intensity value. Efficient modules are also presented for conversion between RGB and HSV color spaces with tunable H and S components in HSV for more flexible color rendering. The design is able to bring out details hidden in shadow regions of the image and preserve the bright parts with adjustable vividness and color shift for improvement of visual quality while maintaining its consistency. It is capable of producing 187.86 million outputs per second (MOPs) on Xilinx’s Virtex II XC2V2000-4ff896 field programmable gate array (FPGA) at a clock frequency of 187.86 MHz. It can process over 179.1 (1024 × 1024) frames per second, which is very suitable for high definition videos, and consumes approximately 70.7% and 76.8% less hardware resource with 127% and 280% performance boost when compared to the designs with machine learning algorithm in [M.Z. Zhang, M.J. Seow, V.K. Asari, A high performance architecture for color image enhancement using a machine learning approach, International Journal of Computational Intelligence Research – Special Issue on Advances in Neural Networks 2(1) (2006) 40–47], and with separated dynamic and contrast enhancements in [H.T. Ngo, M.Z. Zhang, L. Tao, V.K. Asari, Design of a high performance architecture for real-time enhancement of video stream captured in extremely low lighting environment, International Journal of Embedded Systems: Special Issue on Media and Stream Processing, in press], respectively. This approach also provide 83.4 times performance gain with more consistent fidelity in the results compared to some DSP based implementations (256 × 256 frame size) [G.D. Hines, Z. Rahman, D.J. Jobson, G.A. Woodell, DSP implementation of the retinex image enhancement algorithm, visual information processing XIII, in: Proceedings of the SPIE, vol. 5438, 2004, pp. 13–24; G.D. Hines, Z. Rahman, D.J. Jobson, G.A. Woodell, Single-scale retinex using digital signal processors, in: Proceedings of the Global Signal Processing Conference, September 2004, pp. 1–6] under the reflectance-illuminance category of image enhancement models.

Design of a systolic-pipelined architecture for real-time enhancement of color video stream based on an illuminance–reflectance model

A high performance digital architecture for the implementation of a nonlinear image enhancement technique is proposed in this paper. The image enhancement is based on an illuminance–reflectance model which improves the visual quality of digital images and video captured under insufficient or non-uniform lighting conditions. The algorithm shows robust performance with appropriate dynamic range compression, good contrast, accurate and consistent color rendition. The algorithm contains a large number of complex computations and thus it requires specialized hardware implementation for real-time applications. Systolic, pipelined and parallel design techniques are utilized effectively in the proposed FPGA-based architectural design to achieve real-time performance. Approximation techniques are used in the hardware algorithmic design to achieve high throughput. The video enhancement system is implemented using Xilinx's multimedia development board that contains a VirtexII-X2000 FPGA and it is capable of processing approximately 63 Mega-pixels (Mpixels) per second.

2P1-B05 背景色を利用したマーカー色抽出とQ学習による全方位移動型ロボットの制御 : RoboCup小型リーグ

In this paper, we describe two improvements of our robot system made in preparation for competing in the RoboCup Small-Size League. First, we propose a color-recognition method that can be applied under the non-uniform lighting conditions observed in actual games. The basic idea of our proposal is to use the green color of the carpet surrounding the robot as a key for searching a database in order to find proper thresholds in HLS color space. Second, we propose a method for controlling omnidirectional moving robots. The method is based on learning each robot's movement properties. We exploit Q-learning to make the robot perform very simple tasks such as moving straight toward the goal. Our experiments demonstrated the effectiveness of employing these two methods.