For monochrome images, we can perform labeling of all shapes and calculation of basic attributes (area, size, etc.) of regions (primitives) within 1 second, even for complex topographic maps larger than A0 or A1 double-page newspapers.
In our case, we refer to labeled regions as "primitives". Kanji characters and symbols are divided into one or more primitives and analyzed. English uppercase letters and numbers are treated as one character = one primitive.
For monochrome images, there are two types of labeling: 4-connected labeling, which considers regions (primitives) as one when there is a pixel connection in the four directions (up, down, left, right), and 8-connected labeling, which also allows diagonal connections.
For grayscale and color images:
- Labeling based on color reduction processing that statistically minimizes the error with the original image
- Labeling based on simple distance in brightness or color space
- Labeling based on colorization using optimization
- Ability to achieve optimal region division based on the definition of distance between pixels used in the calculations
- Ease of parallelization for accelerated computation
- Ability to be directly applied to videos
The basic attributes that can be calculated using our library are as follows:
- Bounding rectangle Exists between (xmin,ymin) and (xmax,ymax).
- Area
- Perimeter
- Circularity
- Line width
- Absolute maximum length, BR'DTH, ORIENT
Horizontal Feret diameter = xmax-xmin Vertical Feret diameter = ymax-ymin
- Filling regions with background color as noise for regions below a certain area For example, in the case of monochrome, both black and white regions can be labeled, so isolated points or holes can be removed as noise.
- Estimating line width from line drawings
- Estimating the boundaries of strings or paragraphs based on distances and positional relationships between primitives