findContour

Find Contour using NVIDIA NPP

Overview

This sample demonstrates how to detect and extract contours from an image using NVIDIA’s NPP library. It performs unified labeling, label compression, and geometric contour reconstruction on grayscale input images.

Key Concepts

• Unified Framework (UF) Labeling
• Connected Component Analysis
• Image Contour Extraction using CUDA/NPP

Requirements

• Operating System: Linux or Windows
• CPU Architecture: x86_64
• GPU Support: CUDA-enabled GPUs (SM 7.0, 7.2, 7.5, 8.0, and above)
• Toolkit: CUDA Toolkit 11.4 or later
• Libraries: NPP, CMake, and CUDA Runtime

---

Build Instructions

Linux

mkdir build
cd build
cmake ..
make

Windows

mkdir build
cd build
cmake -DCMAKE_GENERATOR_PLATFORM=x64 ..

Then open findContour.sln in Visual Studio (2017 or later) and build the solution.

---

Usage

./findContour

Example Output:

./build/findContour 
Done. Compressed Labels: 274

---

Input & Output

Input Image

• CircuitBoard_2048x1024_8u.raw
• Format: 8-bit unsigned grayscale
• Dimensions: 2048 x 1024

---

Output Visualizations

1. Label Markers (UF-based)

• File: CircuitBoard_LabelMarkersUF_8Way_2048x1024_32u.raw
• Generated by: nppiLabelMarkersUF_8u32u_C1R_Ctx

---

2. Compressed Labels

• File: CircuitBoard_CompressedMarkerLabelsUF_8Way_2048x1024_32u.raw
• Generated by: nppiCompressMarkerLabelsUF_32u_C1IR_Ctx

---

3. Contour Pixels

• File: CircuitBoard_Contours_8Way_2048x1024_8u.raw
• Generated by: nppiCompressedMarkerLabelsUFInfo_32u_C1R_Ctx

---

4. Reconstructed Contours (Ordered Geometry)

• File: CircuitBoard_ContoursReconstructed_8Way_2048x1024_8u.raw
• Generated by: nppiCompressedMarkerLabelsUFContoursOutputGeometryLists_C1R