Pattern Recognition Applications and Methods: 9th International Conference, ICPRAM 2020, Valletta, Malta, February 22–24, 2020, Revised Selected Papers

Preface
Organization
Contents
End to End Deep Neural Network Classifier Design for Universal Sign Recognition
	1 Introduction
	2 Related Work
	3 Methodology
	4 Experimentation and Results
	5 Conclusion and Future Work
	References
MaskADNet: MOTS Based on ADNet
	1 Introduction
	2 Related Work
		2.1 Multi-object Tracking
		2.2 Video Instance Segmentation
		2.3 Segmentation for Tracking
		2.4 Semi-automatic Segmentation
	3 Multi-object Tracking and Segmentation Using MaskADNet
		3.1 Baseline Method
		3.2 MaskADNet
	4 Results and Discussion
		4.1 Dataset Description
		4.2 Evaluation
	5 Conclusion
	References
Dimensionality Reduction and Attention Mechanisms for Extracting Affective State from Sound Spectrograms
	1 Introduction
	2 Related Literature Review
	3 Methodology
		3.1 BoVW Representation Building
		3.2 Choosing Target Variables
		3.3 Temporal Structure Preserving Representation
	4 Experiment Details
		4.1 Dataset Description
		4.2 Experimental Procedure Description
	5 Discussion on Results
	6 Closing Remarks
	References
Efficient Radial Distortion Correction for Planar Motion
	1 Introduction
	2 Related Work
		2.1 Homography Estimation
		2.2 Modelling Radial Distortion
	3 The General Planar Motion Model
	4 Polynomial Solvers
		4.1 A Non-minimal Relaxation (4 Point)
		4.2 Minimal Solver with Known Tilt (2 Point)
	5 Experiments
		5.1 Synthetic Data
		5.2 Numerical Stability
		5.3 Noise Sensitivity
		5.4 Image Stitching
		5.5 Application to Visual Odometry
		5.6 Application to Aerial Imagery
	6 Conclusions
	References
A Preliminary Study on Tree-Top Detection and Deep Learning Classification Using Drone Image Mosaics of Japanese Mixed Forests
	1 Introduction and State of the Art
		1.1 State of Art
	2 Data Gathering, Annotation and Preprocessing
		2.1 Data Acquisition
		2.2 Data Processing and Annotation
		2.3 Challenges in Data and Limitations of This Study
	3 Materials and Methods
		3.1 Interest Region Extraction
		3.2 Tree Top Detection Algorithm
		3.3 Tree Top Classification
	4 Experiments
		4.1 Tree Top Detection
		4.2 Interest Region Extraction
		4.3 Tree Classification
		4.4 Time Considerations
	5 Conclusions
	References
Investigating Similarity Metrics for Convolutional Neural Networks in the Case of Unstructured Pruning
	1 Introduction
	2 Related Work
		2.1 Techniques for DNN Pruning
		2.2 Comparisons Between Pruned and Unpruned DNNs
	3 Tools
		3.1 IMP
		3.2 Similarity Metrics for Neural Networks Data Representations
		3.3 Kernel-Based Metrics
	4 Methods
		4.1 Datasets
		4.2 CNN Architectures and Optimizers Used
	5 Results
		5.1 Test-Set Accuracy
		5.2 Layer-Wise Pruned vs. Unpruned Similarity
		5.3 ResNet_fast
	6 Discussion
		6.1 Takeaways from Results
		6.2 Comparing Output Layers
		6.3 Considerations on the Rotational Invariance of Similarity Metrics for Convolutional Layers
	7 Conclusions and Future Work
	References
Encoding of Indefinite Proximity Data: A Structure Preserving Perspective
	1 Introduction
	2 Non-metric Proximity-Based Learning
		2.1 Notation and Basic Concepts
		2.2 Indefinite Proximities
		2.3 Eigenspectrum Corrections
		2.4 Limitations
	3 Eigenvalue Modification via Nullspace Integrated Shifting
		3.1 Advanced Shift Correction
		3.2 Determination and Approximation of the Shift Parameter
		3.3 Out-of-Sample Extension for New Test Points
		3.4 Structure Preservation of the Eigenspectrum
	4 Experiments
		4.1 Datasets
		4.2 Results
	5 Conclusions
	References
Author Index