Application Benchmarks

Features Download Buy News Support

Application Benchmarks

The following examples illustrate various application categories that SmartSearch can handle. The benchmarks were performed on a 1.7 GHz Pentium 4 with 512 MB RAM running Windows 2000. All the target images are 640x480 unsigned 8-bit. The templates are all 76x47 unsigned 8-bit. In all cases, the entire target image is searched.

The template and target images were chosen to be as similar as possible across all the application categories to minimize complex interaction of conditions, which may confuse the meaning of the benchmarks. All the images were acquired using a standard RS-170 camera with a Coreco Imaging Bandit II frame grabber, with no special lighting other than desk lamps and overhead flourescent lighting.

Category Template Image Parameters Speed (ms)
Simple:
clean image, single target, low noise, no rotation, no scale changes, uniform background

All default values 4
Multiple targets:
Same as simple except multiple targets exist in image. Some objects have same shape but different colors.

Maximum match = 10 4.3
Complex background:
Multiple targets with irrelevant background features.

Maximum match = 10 4.4
Lighting variation:
Target image with lighting conditions significantly different than template. Lighting variation is uniform or gradual.

All default values 4.3
Noisy:
Large amount of random noise. The noise was added in software.

All default values 4.4
Rotation:
Target object in image rotated by 25° relative to trained template.

Rotation range: 0° to 30° at 5° steps 14
Scale:
Target object in image is 60% the size of the template. The scale change was introduced by increasing the distance between the camera and the object.

Scale range: 0.5 to 1 step 0.1
Size: 76x47 12.8
Touch:
Target objects touch but without overlap. Objects have different colors than template.

Method = edge Maximum match = 4 86
Overlap:
Target objects overlap, occluding some features.

Method = edge Maximum match = 3 76
Contrast reversal:
Target objects has a reflective surface, causing pixel values to differ wildly from template, but outline shape of object is same as template.

Method = edge 62

Category	Template	Image	Parameters	Speed (ms)
Simple: clean image, single target, low noise, no rotation, no scale changes, uniform background			All default values	4
Multiple targets: Same as simple except multiple targets exist in image. Some objects have same shape but different colors.			Maximum match = 10	4.3
Complex background: Multiple targets with irrelevant background features.			Maximum match = 10	4.4
Lighting variation: Target image with lighting conditions significantly different than template. Lighting variation is uniform or gradual.			All default values	4.3
Noisy: Large amount of random noise. The noise was added in software.			All default values	4.4
Rotation: Target object in image rotated by 25° relative to trained template.			Rotation range: 0° to 30° at 5° steps	14
Scale: Target object in image is 60% the size of the template. The scale change was introduced by increasing the distance between the camera and the object.			Scale range: 0.5 to 1 step 0.1 Size: 76x47	12.8
Touch: Target objects touch but without overlap. Objects have different colors than template.			Method = edge Maximum match = 4	86
Overlap: Target objects overlap, occluding some features.			Method = edge Maximum match = 3	76
Contrast reversal: Target objects has a reflective surface, causing pixel values to differ wildly from template, but outline shape of object is same as template.			Method = edge	62