MVTec HALCON is implemented for highest performance, e.g., by actively exploiting multi-core computers, NEON, SSE2, AVX, and AVX2, as well as GPU acceleration. The following runtime examples were measured using a byte image of size 640 x 480 on an Intel® Core™ i7-4770 – 3.40 GHz computer. Note: runtime may vary with different input data.
| Shape-based matching (template size: 100 x 100, search area: complete image with 360° rotation): | 0.73 ms |
| Affine transformation (nearest neighbor): | 0.10 ms |
| Sobel edge filter (3 x 3): | 0.08 ms |
| Median (3 x 3): | 0.09 ms |
| Binomial filter (5 x 5): | 0.07 ms |
| Gray opening (3 x 3): | 0.06 ms |
| Binary dilation (50 x 50): | 0.05 ms |
| Binary erosion (50 x 50): | 0.01 ms |
| Threshold operation: | 0.04 ms |
| Subpixel-accurate threshold: | 0.19 ms |
| Feature calculation for 350 objects/blobs; features: “center of gravity” & “number of pixels”: | 0.01 ms |
| Subpixel-accurate measuring of edge positions (search size 50 x 10): | 0.003 ms |
| Fast Fourier transform: | 0.89 ms |
- Automatic GPU Acceleration
For highest performance, HALCON provides an efficient automatic acceleration by optimal usage of the additional computing power of GPUs based on the OpenCL standard. Thereby, more than 80 HALCON operators can be accelerated considerably. Furthermore, HALCON’s deep learning feature also makes use of GPU acceleration.