Print Defect Mapping with Semantic Segmentation
Open accessAbstract
Efficient automated print defect mapping is valuable to the printing industry since such defects directly influence customer-perceived printer quality and manually mapping them is cost-ineffective. Conventional methods consist of complicated and hand-crafted feature engineering techniques, usually targeting only one type of defect. In this paper, we propose the first end-to-end framework to map print defects at pixel level, adopting an approach based on semantic segmentation. Our framework uses Convolutional Neural Networks, specifically DeepLab-v3+, and achieves promising results in the identification of defects in printed images. We use synthetic training data by simulating two types of print defects and a print-scan effect with image processing and computer graphic techniques. Compared with conventional methods, our framework is versatile, allowing two inference strategies, one being near real-time and providing coarser results, and the other focusing on offline processing with more fine-grained detection. Our model is evaluated on a dataset of real printed images.
AI snapshot
Written from the paper’s full text.
TL;DR
The paper proposes an end-to-end semantic segmentation framework based on DeepLab-v3+ to map print defects at the pixel level, trained largely on synthetic dark streaks and color banding with simulated print-scan effects. It reports that no-reference (NR) models generally outperform full-reference (FR) models, with patch-based NR achieving the best mIoU on real data.
Key findings
- NR outperformed FR across both datasets and inference strategies: on semi-synthetic data, resized-image mIoU was 0.32 ± 0.01 for NR vs. 0.25 ± 0.02 for FR, and patch-based was 0.45 ± 0.04 for NR vs. 0.33 ± 0.05 for FR; on real data, resized-image was 0.43 ± 0.02 for NR vs. 0.35 ± 0.03 for FR, and patch-based was 0.49 ± 0.04 for NR vs. 0.37 ± 0.03 for FR.
- Patch-based inference produced finer detection than resized-image inference but was much slower: for example, real-data patch-based NR took 20.0 s/image vs. 1.7 s/image for resized-image NR.
- In a comparison against an intermediate step from Wang et al. [18], the proposed model achieved IoU 0.299 vs. Wang et al.’s 0.183, detecting more streaks including fading ones.
- The framework supports two input modes, NR and FR, and two inference strategies, resized image and patch-based, allowing a tradeoff between speed and quality.
Methods
- Design: an end-to-end deep learning framework for pixel-wise print defect mapping using DeepLab-v3+ semantic segmentation, with NR and FR variants and resized-image and patch-based inference.
- Data: training used 2,949 digital images (documents, photos, slides) with minimum dimensions 1920 × 1080; synthetic defects and print-scan effects were generated online. Evaluation used a semi-synthetic dataset of 103 background images printed/scanned at 600 dpi and a real dataset of 30 scanned images from defective printers, manually annotated.
- Analysis: models were evaluated with mean intersection over union (mIoU) over classes, averaged over four runs; training used multi-class cross-entropy with background weight 0.05, Adam optimizer with initial learning rate 10⁻⁴ and weight decay 10⁻⁵, batch size 5 with gradient accumulation over 10 iterations (effective batch size 50), and a 90%/10% train/validation split.
Limitations
- The paper states that annotation bias is a concern: two human annotators may produce different segmentation masks, and this may be worse for thin print defects, so more realistic evaluation may require metric relaxation in future work.
- The paper states that the framework currently addresses only two defect types, dark streaks and color banding, and that adding more defect types would require simulating them.
- Evident from the text: the real evaluation dataset is small (30 scanned images), and the FR model’s underperformance is hypothesized rather than fully explained, with the authors suggesting concatenation may not explicitly convey that reference and scanned images should be considered separately.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
0 of 19 cited sources checked · 0 supported · 0 flagged
- Findingown data
In preliminary experiments, DeepLab-v3+ outperformed U-Net when segmenting thinner lines.
“In our preliminary experiments, DeepLab-v3+ outperformed U-Net when segmenting thinner lines.”
Quote in the paperSubstantial claim0 of 2 sources checked- cites [14]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [5]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Findingown data
On the real dataset, the No-Reference patch-based method achieved an mIoU of 0.49 ± 0.04.
“Patch-based FR 0.37 ± 0.03 17.8 NR 0.49 ± 0.04”
Quote in the paperSubstantial claim - Findingown data
Patch-based inference is considerably slower than resized image inference.
“However, this advantage comes with the cost of it being considerably slower (Table 1).”
Quote in the paperSubstantial claim - Findingown data
Patch-based inference provides more fine-grained detection compared to the resized image mode.
“Figures 6 and 7 show that the patch-based inference provided more fine-grained detection compared to the resized image mode.”
Quote in the paperSubstantial claim - Findingown data
The No-Reference method outperforms the Full-Reference method on all datasets and inference approaches.
“Table 1 shows that the NR method outperforms FR on all datasets and inference approaches.”
Quote in the paperSubstantial claim
- Findingown data
For the Full-Reference method, concatenating reference and defective images provided better results than using their difference as network input.
“However, we found that concatenation provided better results.”
Quote in the paperNot a substantial claim - Findingown data
The proposed framework achieves promising results in the identification of defects in printed images.
“Our framework uses Convolutional Neural Networks, specifically DeepLab-v3+, and achieves promising results in the identification of defects in printed images.”
Quote in the paperSubstantial claim - Findingown data
Increasing the number of training samples to 11,102 did not give a significant improvement.
“We also verified if we increased the number of training samples to 11 102 would give better results, but no significant improvement was observed.”
Quote in the paperNot a substantial claim - Premisecitation
In large-scale quality control processes, print defect detection still requires human experts in an approach that is repetitive, time-consuming, and prone to human error and subjective evaluation.
“In large-scale quality control processes, print defect detection still requires human experts in an approach that is repetitive, time-consuming, and prone to human error and subjective evaluation [18].”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [18]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Interpretationnone given
The proposed framework does not require calibrating thresholds for different occurrences of streak defects, unlike the method in [21].
“Furthermore, our framework does not require calibrating thresholds for different occurrences of streak defects, as in [21].”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [21]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Interpretationreasoning
The proposed pipeline leverages automatic pixel annotation, resulting in cost-free data in terms of human annotation.
“The proposed pipeline also leverages automatic pixel annotation, resulting in costfree data in terms of human annotation.”
- Also stated in: § 6. Conclusion and Future Works · plain
“Moreover, our method continuously simulates data, making our pipeline cost-free in terms of annotation.”
Quote in the paperSubstantial claim - Also stated in: § 6. Conclusion and Future Works · plain
- Interpretationnone given
The paper presents the first end-to-end framework to map print defects at the pixel level using semantic segmentation.
“In this paper, we propose the first end-to-end framework to map print defects at pixel level, adopting an approach based on semantic segmentation.”
- Also stated in: § 1. Introduction · hedged
“To the best of our knowledge, this is the first work to map print defects at the pixel level.”
- Also stated in: § 6. Conclusion and Future Works · hedged
“To the best of our knowledge, this is the first work to describe a fully-automated pipeline for detecting print defects at the pixel level.”
Quote in the paperSubstantial claim0 of 4 sources checked- cites [16]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [18]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [19]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [20]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Also stated in: § 1. Introduction · hedged
- May combine two claims. 6
- Classified as probably not a substantial claim. 8
May combine two claims.
Findingown dataImageNet-pretrained ResNet-50 achieved similar performance to ResNet-101 while being faster to train.
“The former achieved similar performance while being faster to train.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataAn output stride of 8 gave better results than an output stride of 16, but at the cost of slower training.
“In general, an output stride of 8 gave better results, but at the cost of slower training.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningThe framework adopts two inference strategies, Resized image and Patch-based, taking a tradeoff between speed and quality into account.
“Moreover, it also adopts two inference strategies: Resized image and Patch-based, taking a tradeoff between speed and quality into account.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone givenThe values in Table 1 indicate the average and standard deviation for four runs, to account for randomness.
“Values indicate the average and standard deviation for four runs, to account for randomness.”
Quote in the paperNot a substantial claimMay combine two claims.
InterpretationreasoningThe proposed framework is more general than previous works, working for documents, photos, and combinations of both.
“Conversely, our framework is more general, working for documents, photos, and combinations of both.”
Quote in the paperSubstantial claim0 of 2 sources checked- cites [18]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [21]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
Interpretationown dataThe models can transfer knowledge learned from synthetic data to real print defects, demonstrating applicability for print quality assessment in industry.
“Our results show that our models can transfer the knowledge learned from synthetic data to real print defects, demonstrating its applicability for print quality assessment in industry.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
Interpretationnone givenThe proposed pipeline simulates a print-and-scan effect and artificially creates print artifacts in diverse background images for two types of defects: dark streaks and banding.
“We propose a pipeline for simulating a print-and-scan effect and to artificially create print artifacts in diverse background images for two types of defectsdark streaks and banding.”
Quote in the paperSubstantial claim- Rests on an assumption: Banding and streaks are common to both inkjet and laser electrophotographic printers.
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationDeep convolutional neural networks have achieved remarkable results in many computer vision tasks, including image classification, object detection, and semantic segmentation.
“Since 2012, deep convolutional neural networks (DC-NNs) have achieved remarkable results in many computer vision tasks, including image classification, object detection, and semantic segmentation [4,11,14].”
Quote in the paperNot a substantial claim0 of 3 sources checked- cites [4]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [11]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [14]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
Probably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductown dataManual annotation was not needed for the semi-synthetic dataset because defect locations were recorded during artifact generation.
“Manual annotation was not needed as we recorded defect locations during artifacts generation.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Interpretationnone givenThe framework supports two input modes, Full-Reference and No-Reference, that can be interchangeably used with small changes to its architecture.
“Our framework supports two input modes, Full-Reference (FR) and No-Reference (NR), that can be interchangeably used with small changes to its architecture.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationIn laser printers, the non-uniform line spacing observed in banding defects is mainly caused by fluctuations in the velocity of the optical photoconductor drum.
“In laser printers, the non-uniform line spacing observed in banding defects is mainly caused by fluctuations in the velocity of the optical photoconductor (OPC) drum [6].”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [6]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
PremisecitationDeepLab-v3+ does not require huge datasets, as it was trained on the Pascal VOC 2012 dataset.
“Another advantage of using DeepLab-v3+ architecture is that it does not require huge datasets, as it was trained on the Pascal VOC 2012 dataset [5].”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [5]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationIn inkjet printers, banding is generally caused by small electromechanical or chemical problems in printer components, such as damaged nozzles in the print heads or ink density variation.
“In inkjet printers, banding is generally caused by small electromechanical or chemical problems in the printer components, such as damaged nozzles in the print heads, or ink density variation [3].”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [3]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
PremisecitationPrevious works on print defect detection do not evaluate their approaches in terms of mIoU, and each is evaluated differently, making direct comparisons among works hard.
“Previous works on print defect detection [18,19,21] do not evaluate their approaches in terms of mIoU, and each of them is evaluated differently, making direct comparisons among works hard.”
Quote in the paperSubstantial claim0 of 3 sources checked- cites [18]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [19]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites [21]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
The paper relies on this without support.
The absence of print defects is one of the most significant factors influencing customer satisfaction with a printer product.
“The absence of print defects is one of the most significant factors that influence customer satisfaction when experiencing a printer product.”
The paper relies on this without support.
Banding and streaks are common to both inkjet and laser electrophotographic printers.
“since they are easier to be synthesized, and are both common to inkjet and laser electrophotographic printers.”
The paper relies on this without support.
Print defects are directly linked to the overall quality of printed images.
“Since print defects are directly linked to the overall quality of printed images, we added both sets of metrics.”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Print-scan simulation is described as both the first and last pipeline step
MajorSection 3.1 calls print-scan simulation the last step of synthetic data generation, while Section 3.4 says it happens before defects are added. These descriptions give opposite orders for the same pipeline.
“The last step in our synthetic data pipeline is to imitate the appearance of printing and scanning an image.”
“We first simulate the appearance of printing and scanning effects on digital images.Then, we randomly decide if the image will be defective based on an assigned probability of 0.9.If so, we randomly add the artifacts described in Section 3.1 to it.”
Cites
U-Net: Convolutional Networks for Biomedical Image Segmentation
Ronneberger, O., Fischer, P., & Brox, T. (2015) · Lecture notes in computer science · doi:10.1007/978-3-319-24574-4_28 · Cited by 99,404 · cited 4× · see citing works
“Results on public benchmarks have considerably improved with the use of fully-convolutional DCNNs, such as FCN [11], U-Net [14], and DeepLab [4].”
BackgroundInfluentialAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 1× · see citing works
“We used the Adam optimizer [9] with an initial learning rate of 10, and weight decay of 10.”
MethodFully convolutional networks for semantic segmentation
Long, J., Shelhamer, E., & Darrell, T. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298965 · Cited by 38,834 · cited 2× · see citing works
“Results on public benchmarks have considerably improved with the use of fully-convolutional DCNNs, such as FCN [11], U-Net [14], and DeepLab [4].”
BackgroundChen, L.-C., Papandreou, G., Kokkinos, I., Murphy, K., & Yuille, A. L. (2018) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2017.2699184 · Cited by 22,392 · cited 3× · see citing works
“Results on public benchmarks have considerably improved with the use of fully-convolutional DCNNs, such as FCN [11], U-Net [14], and DeepLab [4].”
BackgroundMethodEncoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
Chen, L.-C., Zhu, Y., Papandreou, G., Schroff, F., & Adam, H. (2018) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-01234-2_49 · Cited by 20,054 · cited 4× · see citing works
“DeepLab-v3+ [5] combines the benefits of two methods: an atrous spatial pyramid pooling (ASPP) module [4] and an encode-decoder structure.”
BackgroundMethodInfluentialDeep Learning in Remote Sensing: A Comprehensive Review and List of Resources
Zhu, X. X., Tuia, D., Mou, L., Xia, G.-S., Zhang, L., Xu, F., & Fraundorfer, F. (2017) · IEEE Geoscience and Remote Sensing Magazine · doi:10.1109/mgrs.2017.2762307 · Cited by 3,400 · cited 1× · see citing works
“Applications of semantic segmentation are diverse, including medical image analysis [14], remote sensing [22] and autonomous driving [15].”
BackgroundFully Convolutional Networks for Semantic SegmentationPreprint
Long, J., Shelhamer, E., & Darrell, T. J. (2014) · arXiv (Cornell University) · doi:10.48550/arxiv.1411.4038 · Cited by 2,490 · see citing works
Convolutional Neural Networks for No-Reference Image Quality Assessment
Kang, L., Ye, P., Li, Y., & Doermann, D. (2014) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2014.224 · Cited by 1,309 · cited 1× · see citing works
“Inspired by works in image quality assessment [2, 7], we also implement two different paradigms for print defect mapping: No-Reference (NR) and Full-Reference (FR).”
MethodDeep Neural Networks for No-Reference and Full-Reference Image Quality Assessment
Bosse, S., Maniry, D., Muller, K.-R., Wiegand, T., & Samek, W. (2017) · IEEE Transactions on Image Processing · doi:10.1109/tip.2017.2760518 · Cited by 1,263 · cited 2× · see citing works
“Inspired by works in image quality assessment [2, 7], we also implement two different paradigms for print defect mapping: No-Reference (NR) and Full-Reference (FR).”
MethodResultBerman, M., Triki, A. R., & Blaschko, M. B. (2018) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2018.00464 · Cited by 1,114 · cited 2× · see citing works
“We tested different dilation values—[1, 1, 1], [1, 2, 1], and [1, 2, 4]— in these 3 layers, achieving the best performances with [1, 2, 1], and [1, 1, 1] in some cases.”
MethodResultU-Net: Convolutional Networks for Biomedical Image SegmentationPreprint
Ronneberger, O., Fischer, P., & Brox, T. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1505.04597 · Cited by 1,079 · see citing works
Deep Neural Networks for No-Reference and Full-Reference Image Quality Assessment
Bosse, S., Maniry, D., Müller, K., Wiegand, T., & Samek, W. (2018) · MPG.PuRe (Max Planck Society) · Cited by 469 · see citing works
Fully Deep Blind Image Quality Predictor
Kim, J., & Lee, S. (2016) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2016.2639328 · Cited by 455 · cited 1× · see citing works
“In contrast to that, having no ground truth image to establish quality criteria makes NR a more challenging problem, which has seen more recent advances with the rise of deep learning [2, 8, 10, 12].”
ResultRTSeg: Real-Time Semantic Segmentation Comparative Study
Siam, M., Gamal, M., Abdel-Razek, M., Yogamani, S., & Jagersand, M. (2018) · 2018 25th IEEE International Conference on Image Processing (ICIP) · doi:10.1109/icip.2018.8451495 · Cited by 112 · cited 1× · see citing works
“onvolutional DCNNs, such as FCN [11], U-Net [14], and DeepLab [4]. Applications of semantic segmentation are diverse, including medical image analysis [14], remote sensing [22] and autonomous driving [15]. In this work, we also employ semantic segmentation techniques, specifically DeepLab-v3+, to the problem of print d…”
MethodBlind Predicting Similar Quality Map for Image Quality Assessment
Pan, D., Shi, P. A., Hou, M., Ying, Z., Fu, S., & Zhang, Y. (2018) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2018.00667 · Cited by 110 · see citing works
No-reference image quality assessment with deep convolutional neural networks
Li, Y., Po, L.-M., Feng, L., & Fang, Y. (2016) · IEEE International Conference on Digital Signal Processing (DSP) · doi:10.1109/icdsp.2016.7868646 · Cited by 77 · cited 1× · see citing works
“In contrast to that, having no ground truth image to establish quality criteria makes NR a more challenging problem, which has seen more recent advances with the rise of deep learning [2, 8, 10, 12].”
ResultOn the Importance of Label Quality for Semantic Segmentation
Zlateski, A., Jaroensri, R., Sharma, P., & Durand, F. (2018) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2018.00160 · Cited by 77 · cited 1× · see citing works
“When it comes to acquiring well-annotated data for training semantic segmentation models, which requires fine-grained label maps, that task becomes even more difficult and costly [23].”
MethodVerikas, A., Lundström, J., Bačauskienė, M., et al. (2011) · Expert Systems with Applications · doi:10.1016/j.eswa.2011.04.035 · Cited by 42 · cited 1× · see citing works
“Some works have proposed automated intelligent systems for assessing print quality [17].”
BackgroundBerman, M., Triki, A. R., & Blaschko, M. B. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1705.08790 · Cited by 39 · see citing works
Image quality assessment using deep convolutional networks
Li, Y., Ye, X., & Li, Y. (2017) · AIP Advances · doi:10.1063/1.5010804 · Cited by 28 · see citing works
Banding Characterization for Inkjet Printing.
Briggs, J. C., Murphy, M., & Pan, Y. (2000) · Cited by 24 · see citing works
Real-time Defect Detection Method for Printed Images Based on Grayscale and Gradient Differences
Yangping, W., Shaowei, X., Zhengping, Z., et al. (2018) · Journal of Engineering Science and Technology Review · doi:10.25103/jestr.111.22 · Cited by 18 · cited 2× · see citing works
“[20] proposed a sys-tem based on grayscale and gradient differences for template matching.”
BackgroundHandling Inter-Annotator Agreement for Automated Skin Lesion SegmentationPreprint
Ribeiro, V., Avila, S., & Junior, E. A. D. V. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.02415 · Cited by 17 · cited 1× · see citing works
“This problem is recurrent in other applications of semantic segmentation [13].”
BackgroundLocal Defect Detection and Print Quality Assessment
Wang, J., Nelson, T., Jessome, R., et al. (2016) · Electronic Imaging · doi:10.2352/issn.2470-1173.2016.13.iqsp-207 · Cited by 8 · see citing works
Real-Time Print Quality Diagnostics
Xiao, Z., Nguyen, M., Maggard, E., et al. (2017) · Electronic Imaging · doi:10.2352/issn.2470-1173.2017.12.iqsp-239 · Cited by 5 · cited 3× · see citing works
“Previous works on print defect detection [18, 19, 21] do not evaluate their approaches in terms of mIoU, and each of them is evaluated differently, making direct comparisons among works hard.”
BackgroundMethodAssessment of Color Fringe on Scanned Images
Jang, S. K., & Kim, C.-W. (2012) · Journal of Imaging Science and Technology · doi:10.2352/j.imagingsci.technol.2012.56.1.010505 · Cited by 3 · see citing works
Block window method with logistic regression algorithm for streak detection
Zhang, R., Maggard, E., Jessome, R., et al. (2019) · Electronic Imaging · doi:10.2352/issn.2470-1173.2019.10.iqsp-300 · Cited by 3 · cited 4× · see citing works
“[21] adopted a similar approach to detect light streaks in printer test images.”
BackgroundMethodInfluentialSimulation of Print Quality Defects
Jang, W., & Allebach, J. P. (2005) · Journal of Imaging Science and Technology · doi:10.2352/j.imagingsci.technol.2005.49.1.art00002
Flagged first, then by citations
Research works printed but not linked: 1
- spivakovsky, 0, Image defect detection, May 10 2018, US Patent
Cited by
Sorted by citations
Joining Spatial Deformable Convolution and a Dense Feature Pyramid for Surface Defect Detection
Huang, Z., Zhang, C., Ge, L., Chen, Z., Lü, K., & Wu, C. (2024) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2024.3370962 · Cited by 89 · see citing works
“Although detection-based surface defect detection methods have significantly improved in recent years, most existing methods [40], [41], [42], [43], [44] are incapable of accurately detecting surface defects with complicated shapes and multiscale sizes.”
MethodAttentive Boundary-Aware Fusion for Defect Semantic Segmentation Using Transformer
Yeung, C.-C., & Lam, K. (2023) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2023.3271723 · Cited by 48 · see citing works
“It has been applied to various industrial applications [23], [24], [25].”
BackgroundSurvey on industrial defect detection with deep learning
罗, 东., 蔡, 雨., 杨, 子., 章, 哲., 周, 瑜., & 白, 翔. (2022) · Scientia Sinica Informationis · doi:10.1360/ssi-2021-0336 · Cited by 43 · see citing works
Shao, L., Zhang, E., Duan, J., & Ma, Q. (2023) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2023.106369 · Cited by 36 · see citing works
Li, D., Li, J., Fan, Y., Lu, G., Ge, J., & Liu, X. (2022) · Expert Systems with Applications · doi:10.1016/j.eswa.2022.117372 · Cited by 23 · see citing works
Advancing Ultrasonic Defect Detection in High-Speed Wheels via UT-YOLO
Zhang, Q., Peng, J.-P., Tian, K., Wang, A., Li, J.-L., & Gao, X.-R. (2024) · Italian National Conference on Sensors · doi:10.3390/s24051555 · Cited by 18 · see citing works
An Efficient Printing Defect Detection Based on YOLOv5-DCN-LSK
Liu, J., Cai, Z., He, K., Huang, C., Lin, X., Liu, Z., Li, Z., & Chen, M. (2024) · Sensors · doi:10.3390/s24237429 · Cited by 8 · see citing works
Investigating Shift Equivalence of Convolutional Neural Networks in Industrial Defect Segmentation
Qu, Z., Tao, X., Shen, F., Zhang, Z., & Li, T. (2023) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2023.3325514 · Cited by 8 · see citing works
“To address the limited sample issue in print defect datasets, Valente et al. [18] utilized computer graphics techniques to synthesize new training samples at the pixel level, resulting in commendable segmentation performance using DeepLabv3 + .”
MethodA two-stage defect detection method for unevenly illuminated self-adhesive printed materials
Peng, G., Song, T., Cao, S., Zhou, B., & Jiang, Q. Y. (2024) · Scientific Reports · doi:10.1038/s41598-024-71514-z · Cited by 4 · see citing works
“Unlike traditional methods, deep learning-based ones primarily utilize convolutional neural networks (CNN) to extract image defect features 21 , leveraging their unique advantages in mining defect information and demonstrating more robust recognition performance for surface defects 22–27 .”
MethodDifferential Enhanced Siamese Segmentation Network for Printed Label Defect Detection
Li, D., Li, Y., Li, J., et al. (2023) · Proceedings - International Conference on Image Processing · doi:10.1109/icip49359.2023.10222759 · Cited by 4 · see citing works
“Segmentation networks [7, 8] predict defects through pixel-wise segmentation to better obtains the shape of defects.”
BackgroundTriple-Stream Siamese Segmentation Network for Printed Label Defect Detection
Li, D., Li, Y., Ma, X., Li, J., & Lu, G. (2024) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2024.3472770 · Cited by 2 · see citing works
Application of Artificial Intelligence in Printing Industry: Systematic Review
Masod, M. Y., & Zakaria, S. F. (2024) · doi:10.1109/icspc63060.2024.10861924 · Cited by 1 · see citing works
VERIFIKASI DOKUMEN CETAK MENGGUNAKAN METODE EDGE DETECTION-GLCM DAN K-MEAN CLUSTERING
Kurniati, F. T., & Huizen, R. R. (2020) · Dinamik · doi:10.35315/dinamik.v25i2.8188
“Ciri yang terdapat pada dokumen cetak bisa berasal dari kecacatan saat proses cetak (print defect) [7], [10].”
BackgroundBDRBNet: an efficient network for semantic segmentation of industrial point clouds
Jian, Z., gao, H., Guo, Z., et al. (2024) · International Conference on Control, Robotics and Intelligent System (CCRIS) · doi:10.1117/12.3050196
“The use of point clouds for industrial defect detection has several advantages: First, compared to traditional manual inspection, this computer vision-based technology 2–6 is simple to deploy, low in cost, and easy to maintain.”
ASFormer: attentive semantic feature fusion transformer for pixel-level defect detection
Zhu, Q., Hu, H., Liu, T., et al. (2024) · Measurement Science and Technology · doi:10.1088/1361-6501/ad9bcf
“Unlike traditional defect detection methods, pixel-level defect detection classifies each pixel as either defective or non-defective, a technique that has been applied in various industrial settings [29, 30].”
Exploring the scope of deep neural network to predict print densitometry from scanner data
Debnath, S., & Chatterjee, A. (2025) · Neural Computing and Applications · doi:10.1007/s00521-025-11313-3
- Authors
- Valente, A. C., Wada, C., Neves, D., Neves, D., Perez, F. V. M., Megeto, G. A. S., Cascone, M. H., Gomes, O., & Lin, Q.
- Affiliations
- Instituto de Pesquisas Eldorado (BR)
- Hewlett-Packard (United States) (US)
- Venue
- 2020 IEEE Winter Conference on Applications of Computer Vision (WACV)
- Publisher
- IEEE
- Published
- 2020-03
- Pages
- 3540-3548
- ISSN
- 2472-6737 (ISSN-L)
- OpenAlex
- W3025011308
- Semantic Scholar
- 71e9ed7354b2cda5e8960fd3337c0c78b8e5150e
- Cited by
- 16
- References
- 26
- Citations by year
- 2025: 1 · 2024: 8 · 2023: 4 · 2022: 2 · 2020: 1
- FWCI
- 2.49
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE Winter Conference on Applications of Computer Vision (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Hewlett-Packard Development Company
- Topics
- Physical Sciences › Engineering › Industrial and Manufacturing Engineering › Industrial Vision Systems and Defect Detection
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Engineering › Civil and Structural Engineering › Infrastructure Maintenance and Monitoring
- Keywords
- semantic segmentation, printed images, DeepLabV3+, convolutional neural networks, synthetic training data, end-to-end detection, printing quality inspection
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-17.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-17.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
