Improved Binarization Algorithm of DM Code Image under Complex Illumination

Façade image recognition under complex illumination conditions is crucial for various applications, including urban three-dimensional modeling and building identification. Existing methods relying solely on Red-Green-Blue (RGB) images are prone to texture ambiguity in complex illumination environments. Furthermore, façades display varying orientations and camera viewing angles, resulting in performance issues within the RGB color space. In this study, we introduce an illumination-invariant façade recognition network (I2-FaçadeNet) that leverages sparsely gated multi-color space experts for enhanced façade image recognition in challenging illumination environments. First, RGB façade images are converted into multi-color spaces to eliminate the ambiguous texture in complex illumination. Second, we train expert networks using separate channels of multi-color spaces. Finally, a sparsely gated mechanism is introduced to manage the expert networks, enabling dynamic activation of expert networks and the merging of results. Experimental evaluations leveraging both the International Society for Photogrammetry and Remote Sensing benchmark data sets and the Shenzhen data sets reveal that our proposed I2 -FaçadeNet surpasses various depths of ResNet in façade recognition under complex illumination conditions. Specifically, the classification accuracy for poorly illuminated façades in Zurich improves by nearly 8%, while the accuracy for over-illuminated areas in Shenzhen increases by approximately 3%. Moreover, ablation studies conducted on façade images with complex illumination indicate that compared to traditional RGB-based ResNet, the proposed network achieves an accuracy improvement of 3% to 4% up to 100% for overexposed images and an accuracy improvement of 3% to 10% for underexposed images.

In close-range photogrammetry, circular-coded targets (CCTs) are a reliable method to solve the issue of image correspondence. Currently, the identification methods for CCTs are very mature, but complex illumination conditions are still a key factor restricting identification. This article proposes an adaptive homomorphic filtering (AHF) algorithm to solve this issue, utilizing homomorphic filtering (HF) to eliminate the influence of uneven illumination. However, HF parameters vary with different lighting types. We use a genetic algorithm (GA) to carry out global optimization and take the identification result as the objective function to realize automatic parameter adjustment. This is different from the optimization strategy of traditional adaptive image enhancement methods, so the most significant advantage of the proposed algorithm lies in its automation and universality, i.e., users only need to input photos without considering the type of lighting conditions. As a preprocessing algorithm, we conducted experiments combining advanced commercial photogrammetric software and traditional identification methods, respectively. We cast stripe- and lattice-structured light to create complex lighting conditions, including uneven lighting, dense shadow areas, and elliptical light spots. Experiments showed that our algorithm significantly improves the robustness and accuracy of CCT identification methods under complex lighting conditions. Given the perfect performance under stripe-structured light, this algorithm can provide a new idea for the fusion of close-range photogrammetry and structured light. This algorithm helps to improve the quality and accuracy of photogrammetry and even helps to improve the decision making and planning process of photogrammetry.

We improved classic retinal modeling to alleviate the adverse effect of complex illumination on face recognition and extracted robust image features. Our improvements on classic retinal modeling included three aspects. First, a combined filtering scheme was applied to simulate functions of horizontal and amacrine cells for accurate local illumination estimation. Second, we developed an optimal threshold method for illumination classification. Finally, we proposed an adaptive factor acquisition model based on the arctangent function. Experimental results on the combined Yale B; the Carnegie Mellon University poses, illumination, and expression; and the Labeled Face Parts in the Wild databases show that the proposed method can effectively alleviate illumination difference of images under complex illumination conditions, which is helpful for improving the accuracy of face recognition and that of facial feature point detection.

Image binarization is divided into global algorithm and local algorithm. Global binarization algorithms have a problem to describe objects that have similar brightness with a single threshold. Local binarization algorithms make boundary lines because these algorithms split the image into a specific size of blocks. Therefore, in this paper, we propose a binarization method to complement these problems. The proposed method uses triangular fuzzy membership function to classify the image into obvious regions and ambiguous regions. Obvious regions are binarized by using global binarization algorithm. Whereas ambiguous regions are binarized by using improved local algorithm. Experimental results show the proposed method binarizes the image with less information loss. Moreover, binarized image describes the object in more detail than global binarization methods and more natural than local binarization method.

Accurate and effective water leakage detection in tunnel linings under complex illumination conditions is crucial for ensuring operational safety. This paper proposed a method for detecting water leakage based on infrared vision and an enhanced U-shaped network (UNet) architecture. Initially, the lightweight Azure Kinect sensor was employed to capture both infrared and RGB images of tunnel linings under different illumination conditions. Three datasets: the RGB image under normal illumination (NRGB), the infrared image under normal illumination (NIR), and the infrared image under low illumination (LIR) were created. Subsequently, an enhanced UNet architecture (UKAN) was introduced for water leakage segmentation in tunnel linings. Moreover, Score-weighted and Layer-wise class activation mapping (CAM) were adopted to analyze the model’s decision-making process. Experimental results revealed that the average detection accuracy for the NIR exceeded that of the NRGB by 2.4%, while the LIR was higher by 1.43%. The proposed UKAN architecture achieved optimal mean intersection over union scores of 84.1%, 86.5%, and 85.7% on NRGB, NIR, and LIR. Additionally, visual interpretability analysis showed that the UKAN’s learning process was smooth, gradually shifting its focus from low-level features to target regions. This methodology facilitates efficient and accurate water leakage detection in tunnel linings under varying illumination conditions.

We present a non-rigid registration technique that achieves spatial, photometric, and visibility accuracy. It lets us photo-realistically augment 3D deformable surfaces under complex illumination conditions and in spite of severe occlusions. There are many approaches that address some of these issues but very few that simultaneously handle all of them as we do. We use triangulated meshes to model the geometry and introduce explicit visibility maps as well as separate illumination parameters for each mesh vertex. We cast our registration problem in an Expectation Maximization framework that allows robust and fully automated operation. It provides explicit illumination and occlusion models that can be used for rendering purposes.

It is a challenging task to improve the performance of face recognition under complex illumination conditions. Illumination estimation-based illumination invariant extraction is widely used to alleviate the adverse effects of illumination variation on face recognition. Most existing methods only used slowly changing characteristics of lighting to achieve illumination estimation, thus resulting in inaccurate illumination estimation and illumination invariant extraction under complex illumination conditions. To alleviate this issue, on the basis of the Lambertian reflectance model, we propose an innovative method of directional illumination estimation to extract directional illumination invariant sets from a facial image. The directional illumination invariant sets not only better preserve essential features of the face, but also largely reduce adverse effects of rapid light changes. Moreover, we propose a multilevel matching metric for category classification by using an inner product measure and residual matching. Experimental results on Yale $\text{B}^{+}$ , CAS-PEAL-R1, uncontrolled and AR face databases validate that the proposed method can effectively improve the accuracy of face recognition under complex illumination conditions.

Environment awareness technology is one of the most critical technologies for autonomous vehicles, and traffic sign recognition is an important branch of environment awareness system, which has important research significance for ensuring traffic safety. To solve the problems of easy omission and inaccurate positioning for traffic sign detection under complex illumination conditions, there are two main innovations in this paper: firstly, an adaptive image enhancement algorithm was proposed to improve the image quality under complex illumination conditions; Secondly, a novel and lightweight attention block named Feature difference (FD) model was proposed to detect and recognize traffic signs. Unlike state-of-the-art attention models, which utilizes the difference between two feature maps to generate the attention mask. In this work, the single-stage target detection algorithm SSD was selected as the basic network, the backbone network was set as ResNet and VGG respectively and FD module was added. A large number of experiments were carried out to evaluate the optimization effect of FD module. The following conclusions can be drawn: image enhancement algorithms can provide better image samples; The FD model can be integrated into most of CNNs just by adding some shortcut connections, which does not introduce any additional parameters and layers; The average detection accuracy with FD module is 1.80% higher and the average recall rate was 1.51% higher than that without FD module, which has no great influence on the running speed. The FD module has better enhancement effect for dark image detection; The detection accuracy of the FD module is slightly better than other attention modules including BAM, CBAM and SE; The FD model can be used to evaluate the effect of the convolutional operation on features, which could help to find convolutional layers that have little effect on the accuracy of recognition for the network pruning.

Perception in trellised orchards is often challenged by dense canopy occlusion and overhead plastic coverings, which cause pronounced variations in sky visibility at row terminals. Accurately recognizing row terminals, including both row head and row tail positions, is therefore essential for understanding orchard row structures. This study presents SkySeg-Net, a sky segmentation-based framework for row-terminal recognition in trellised orchards. SkySeg-Net is built on an enhanced multi-scale U-Net architecture and employs ResNeSt residual split-attention blocks as the backbone. To improve feature discrimination under complex illumination and occlusion conditions, the Convolutional Block Attention Module (CBAM) is integrated into the downsampling path, while a Pyramid Pooling Module (PPM) is introduced during upsampling to strengthen multi-scale contextual representation. Sky regions are segmented from both front-view and rear-view camera images, and a hierarchical threshold-based pixel-sum analysis is applied to infer row-terminal locations based on sky-region distribution patterns. To support a comprehensive evaluation, a dedicated trellised vineyard dataset was constructed, featuring front-view and rear-view images and covering three representative grapevine growth stages (BBCH 69–71, 73–77, and 79–89). Experimental results show that SkySeg-Net achieves an mIoU of 91.21% and an mPA of 94.82% for sky segmentation, with a row-terminal recognition accuracy exceeding 98.17% across all growth stages. These results demonstrate that SkySeg-Net provides a robust and reliable visual perception approach for row-terminal recognition in trellised orchard environments.

In this paper, an automatic number plate recognition approach with low computational load has been proposed to detect licence plate area using character features. In the preprocessing step, unlike classical Sauvola method the output is weighted according to the pixel luminance values, and therefore dark regions are eliminated from the detection. After preprocessing step, regions that cannot show a character property are eliminated using connected component analysis, and then character regions are detected utilizing horizontal projection. Experimental results show that proposed method works faster and gives better detection performance in complex background, variable illumination, distance and inclination conditions.

Urban trees provide important ecosystem services to improve cities’ liveability and sustainability. Leaf chlorophyll content (Cab) estimation by remote sensing can help monitor tree health efficiently. However, the Cab retrieval of urban trees is challenging because of the complex canopy structures, backgrounds, and illuminations conditions. This paper proposed an automatic method for partitioning sunlit/shaded pixels and removal of bright-specular/dark-hole and background pixels. In addition, we proposed a new index, the Urban Tree Chlorophyll Index (UTCI), defined as UTCI=(ρ709-ρ697)/(ρ709-ρ686), based on the simulated hyperspectral images of urban tree using radiative transfer model. This proposed UTCI index outperforms existing narrow-band indices (NBIs) for estimating Cab for complex canopy structures, backgrounds, and illumination conditions evaluated using simulated hyperspectral images. The advantage of the UTCI was also demonstrated when applied to UAV hyperspectral images validated with direct field foliar measurements of Cab. It surpasses existing NBIs, demonstrating a moderate correlation (R2 = 0.34) with Cab under varying irradiance and a strong correlation (R2 = 0.62) with Cab under stable diffuse illumination. This study, for the first time, extensively investigated NBIs for Cab estimation of urban trees from UAV hyperspectral images, providing a theoretical and operational basis for future monitoring of Cab in the management of urban trees. This new method can be potentially applied to other vegetation types with complex canopy structures, backgrounds, and illuminations conditions.

Detecting text in video or natural scene image is quite challenging due to the complex background, various fonts and illumination conditions. The preprocessing period, which suppresses the nontext areas so as to highlight the text areas, is the basis for further text detection. In this paper, a novel graph-based background suppression method for scene text detection is proposed. Considering each pixel as a node in the graph, our approach incorporates pixel-level and context-level features into a graph. Various factors contribute to the unary and pair wise cost function which is optimized via max-flow/min-cut algorithm [16] to get a binary image whose nontext pixels are suppressed so that text pixels are highlighted. Furthermore, the proposed background suppression method could be easily combined with other detection methods to improve the performance. Experimental results on ICDAR 2011 competition dataset show promising performance.

High dynamic range (HDR) displays with dual-panels are one type of displays that can provide HDR content. These are composed of a white backlight panel and a colour LCD panel. Local dimming algorithms are used to control the backlight panel in order to reproduce content with high dynamic range and contrast at a high fidelity. However, existing local dimming algorithms usually process low dynamic range (LDR) images, which are not suitable for processing HDR images. In addition, these methods use hand-crafted features to estimate the backlight values, which may not be suitable for many kind of images. In this work, a novel deep learning based local dimming method is proposed for rendering HDR images on dual-panel HDR displays. The method uses a Convolutional Neural Network (CNN) to directly predict backlight values, using as input the HDR image that is to be displayed. The model is designed and trained via a controllable power parameter that allows a user to trade off between power and quality. The proposed method is evaluated against seven other methods on a test set of 105 HDR images, using a variety of quantitative quality metrics. Results demonstrate improved display quality and better power consumption when using the proposed method compared to the best alternatives.

Substation equipment carries critical operational information. For the intelligent monitoring of power systems, the deployment of camera-equipped wheeled robots and unmanned aerial vehicles for image acquisition and real-time detection of substation equipment represents an essential routine inspection methodology. While modern detection algorithms have achieved widespread adoption in power industry applications, state-of-the-art methods still face significant performance degradation when handling dense small-target detection tasks in complex field environments. To address these limitations, this paper proposes a novel spatial detail-guided and clue-screening network (SDG-CSNet), including an efficient clue-scanning module (ECSM), visual detail retention (VDR), and a detail-guided path aggregation network (DG-PAN). To tackle the degradation of image quality induced by complex illumination conditions, the input images undergo edge detail smoothing via preprocessing, while the ECSM utilizes a rescreening mechanism to effectively integrate local connectivity with global receptive fields for feature reconstruction. Addressing the characteristics of densely distributed and small-sized objects, VDR employs pixel-adaptive downsampling strategies to maintain high-frequency information. Moreover, DG-PAN implements delayed multi-scale fusion of shallow-level features, thereby guiding feature maps to adjust dynamically. Experiments demonstrate that SDG-CSNet achieves a state-of-the-art performance with 57.3% mAP. The detection speed, model size, and computation are 59.9 frames per second, 6.25 M, and 14.1 G. SDG-CSNet demonstrates superior robustness in low-light and motion-blur conditions. This work provides a practical and efficient solution for substation equipment detecting applications.

Under maize-soybean rotation systems, weeds and crops at the seedling stage in dryland fields exhibit high similarity in morphological structure, scale distribution, and spatial arrangement. In addition, complex illumination conditions, occlusion, and background interference further complicate accurate weed discrimination. To address these challenges, this study proposes an improved YOLOv11n-based weed detection method for seedling-stage crops under dryland rotation conditions, aiming to enhance detection accuracy and robustness in UAV-acquired field images. Three key improvements were introduced to enhance model performance: (1) the incorporation of Dynamic Convolution (DynamicConv) to adaptively strengthen feature representation for weeds with varying morphologies and scales in low-altitude remote sensing imagery; (2) the design of a SlimNeck lightweight feature fusion architecture to improve multi-scale feature propagation efficiency while reducing computational cost; (3) the cascaded group attention mechanism (CGA) is integrated into the C2PSA module, thereby improving discrimination capability under complex background conditions. These results represent consistent improvements over baseline models, including YOLOv5, YOLOv6, YOLOv8, YOLOv11, and YOLOv12. Specifically, detection performance for broadleaf weeds and Poaceae weeds reached mAP@0.5 values of 87.2% and 73.9%, respectively. Overall, the proposed method demonstrates superior detection accuracy and stability for seedling-stage weed identification under rotation conditions, providing reliable technical support for variable-rate herbicide application and precision field management.