Kittipong Tapyou

Comparative Evaluation of Point Cloud Upsampling Networks for Real-World LiDAR Terrain Reconstruction

Tue, 25 Aug 2026 00:00:00 +0000

Abstract

Airborne LiDAR point clouds represent terrain but are often sparse, noisy, and unevenly distributed, limiting their use in surface reconstruction and geospatial analysis. This study benchmarks four learning-based point cloud upsampling architectures, PU-Net, MPU, PU-GCN, and PU-Transformer, using real-world terrain data from the U.S. Geological Survey 3D Elevation Program (3DEP). A unified Farthest Point Sampling (FPS)-based preprocessing pipeline generates input and ground-truth patch pairs. Each model is trained with a fixed 4× upsampling ratio and input patch sizes of 128, 256, 512, and 1024 points. Performance is assessed on a held-out terrain sub-region using Chamfer Distance, Hausdorff Distance, Point-to-Surface Distance, Earth Mover’s Distance, and the Normalized Uniformity Coefficient. MPU achieves the best mean surface fidelity and point uniformity, obtaining the lowest Chamfer Distance, Point-to-Surface Distance, and Normalized Uniformity Coefficient at 1024 input points. PU-GCN achieves the lowest Hausdorff Distance and Earth Mover’s Distance at 512 input points. PU-Transformer remains competitive, whereas PU-Net produces consistently larger errors and does not benefit reliably from denser inputs. The three more recent architectures generally improve as patch capacity increases, indicating that richer local context improves terrain reconstruction. Overall, MPU and PU-GCN are the most suitable approaches for real-world LiDAR terrain point cloud upsampling tasks.

Patch-Based Bézier Surface Upsampling of Gridded Rainfall Data in the Chao Phraya River Basin, Thailand

Tue, 25 Aug 2026 00:00:00 +0000

Abstract

High-resolution rainfall data are essential for basin-scale hydrological analysis, but long-term rainfall products are often available only at coarse spatial resolutions. This study proposes a patch-based upsampling method using Bézier surfaces to enhance the spatial resolution of gridded rainfall data over the Chao Phraya River Basin, Thailand. Daily rainfall observations from 79 Thai Meteorological Department stations during 1981 – 2021 were quality controlled and interpolated using Inverse Distance Weighting (IDW) to generate 15-arc-minute rainfall grids. Each coarse grid was subdivided into rectangular patches based on local rainfall peak locations. A tensor-product Bézier surface was fitted to each patch and sampled to produce a 5-arc-minute rainfall field. The proposed method was evaluated using spatial comparisons and monthly basin-average rainfall over the Chao Phraya River Basin. The upsampled fields preserved the broad spatial rainfall pattern of the coarse input while providing smoother rainfall gradients at the target resolution. Monthly basin-average rainfall from the Bézier -upsampled fields showed near-perfect agreement with the directly interpolated 5-arc-minute IDW reference field, with R^2=0.999 for 1981 – 2021. Strong agreement was also obtained with the T12 dataset (R^2=0.981) and K10 dataset (R^2=0.844), whereas agreement with the Thiessen polygon dataset was moderate (R^2=0.663). The results

Variable Thresholding

Sun, 02 Nov 2025 00:00:00 +0000

Variable Thresholding

This lab implements adaptive thresholding to handle uneven illumination in grayscale images. Instead of one global threshold, the image is divided into smaller partitions, and each part gets its own local Otsu threshold. Result: Local thresholds adapt to illumination, correctly separating dark circular objects from the background much better than global thresholding.

Hole Filling for 3D Point Clouds via Bézier Curve with C1-Continuity Generation

Sat, 01 Nov 2025 00:00:00 +0000

We reconstruct incomplete point clouds using C1-continuous cubic Bézier curves for accurate, smooth surface completion.

A Novel Triangulation-Based Method for Hole Filling in 3D Point Clouds Using Approximation Curve Fitting

Fri, 31 Oct 2025 00:00:00 +0000

We reconstruct missing point-cloud regions using triangulated control-point curves and uniform sampling for accurate surface completion.

Image Restoration

Tue, 28 Oct 2025 00:00:00 +0000

Blurring Function and Restoration

Simulates linear motion blur and performs restoration in the frequency domain.

Simulates sequential horizontal and vertical motion blurs (T₁, T₂).
Restores blurred images by dividing Fourier-transformed images with known H(u, v).

Observation:
Effective restoration achieved when motion parameters (a, b, T) are known and noise is negligible.

Morphological Operation

Tue, 21 Oct 2025 00:00:00 +0000

Morphological Operations

Implements connected-component labeling using geodesic dilation:

Invert the binary image to make objects white.
Select an unlabeled white pixel as a seed.
Perform geodesic dilation constrained by the original mask.
Label each connected region iteratively.

Hough Transform

Tue, 14 Oct 2025 00:00:00 +0000

Hough Transform

Manual implementation for:

Line detection: Based on Sobel edge maps, ρ–θ parameterization, and local maxima in Hough space.

Circle detection: Uses tangent direction estimation to compute (a, b, r).

Result:
Successfully detects multiple lines and circles in complex images such as clocks and coins by iterative voting and duplicate suppression.

Spatial Filtering

Tue, 23 Sep 2025 00:00:00 +0000

Spatial Filtering

Implements:

Low-pass filters: Mean and Gaussian
High-pass filters: Laplacian and Laplacian of Gaussian (LoG)
Includes a custom convolution with edge-padding (ConvPadded).

Observation:

Mean filter increases blur with kernel size.
Gaussian filter maintains more structure.
LoG detects sharper edges compared to Laplacian alone.

Histogram Equalization

Tue, 16 Sep 2025 00:00:00 +0000

Histogram Equalization

Steps:

Compute histogram and cumulative distribution function (CDF).
Normalize CDF and map it to new intensity values (HistEqualize function).
Compare manual results with Mathematica’s HistogramTransform.

Result:
Custom implementation produces contrast enhancement nearly identical to the built-in method, with well-distributed intensity levels.

Thresholding

Wed, 20 Aug 2025 00:00:00 +0000

Maximum Normal Line and Average Intensities Method

Location

Wed, 08 May 2024 00:00:00 +0000

Professor Dr von Igelfeld's mailing and office addresses at the Institute of Romance Philology.

Office Hours

Wed, 08 May 2024 00:00:00 +0000

Schedule and location for Professor Dr von Igelfeld's office hours.

Smart School Attendance System using Face Recognition with Near Optimal Imaging

Tue, 27 Jul 2021 00:00:00 +0000

Face-recognition attendance system using web and IoT improves accuracy by optimizing student position and distance..

Lectures at the Villa of Reduced Circumstances

Mon, 10 Mar 2003 00:00:00 +0000

This graduate course presents classical results in Romance philology.

Portugese Irregular Verbs

Wed, 01 Jan 1997 00:00:00 +0000

This book discusses Portugese irregular verbs in great details.

Profanity in Early Irish

Sat, 16 Jun 1990 00:00:00 +0000

This undergraduate course discusses profanity in Early Irish.