# IKOSA AI Output Parameters

This page lists the standard center of mass of the output. The center of mass is calculated over the object's shape. The underlying intensities do not affect this value.

The object center serves a dual purpose: it places the object ID in IKOSA AI visualizations and calculates the nearest neighbor distances.

**Important: **For objects with holes or thin structures with a strong overall curvature, the center may not be on the object’s surface.

## is_at_border

**Type: ****bool**

**Example:** True / False

Flag if the found object is at the image border or on the border of a used Region of Interest (ROI). As these objects might not be fully represented within the image, the calculated parameters (area, intensities, etc.) might not correlate to the object's true values, negatively affecting statistical values calculated on the whole set.

**Important: **This value enables the user to filter the output for only objects guaranteed to be within the analysis area.

## has_holes

**Type: ****bool**

**Example:** True / False

Flag if the found object contains any holes. Besides serving as a detector for objects containing holes, this can also serve as a filter for reliable data in the output. Experience has shown that the erroneous prediction of holes within objects is often seen in applications lacking sufficient training data. Filtering for objects without holes can exclude potentially divergent outputs in your analysis.

**Important**: The flag is sensitive to even 1-pixel holes.

## is_recombined

**Type: ****bool**

**Example:** True / False

Flag if the object has been recombined at a tile border during calculation. As the internal calculation divides large images into tiles (default size 2048 x 2048 Px), objects at these tile borders might be divided and detected twice. IKOSA AI identifies split objects and recombines them during the analysis process so that most parameters are calculated as if the object has never been divided.

However, there are parameters that, in certain situations, might be affected by such a split:

**Outgrowth:**

Outgrowth is calculated on each tile and recombined if the split is detected. This can lead to incorrectly missing shifted outgrowth positions:

Usual Case - correct prediction:

Missing Outgrowth overspill:

Incorrect touching point / Incomplete Outgrowth:

The flag can help filter for erroneously calculated outgrowths if the object was split. Errors cannot be detected that way because objects are near the border.

**Nearest Neighbor:**

When calculating the nearest neighbor of a split object, only the two previous nearest neighbors of the two object parts are considered as new nearest neighbors. In certain object shapes and positions, a third object might be closer to the recombined object center. The flag can be used to ensure that only objects entering statistical analysis have a guaranteed correct nearest neighbor.

## object_area / outgrowth_area / bbox_area

**Type: ****float**

**Example:** 10532.0

Number of pixels within the object borders/outgrowth borders/bounding box on the original scale.

## distance_nearest_neighbor

**Type: ****float**

**Example:** 132.52

Distance between the object center of the current object to the object center of the adjacent object in Pixels.

## area_ratio

**Type: ****float**** [0 - 100]**

**Example:** 80.5

The ratio of the object area to the area of the whole image or ROI in percent.

## perimeter

**Type: ****float**

**Example:** 19753.52

Length of the object perimeter. The perimeter is measured as the circumference around the center position of the outermost pixels of the object. To make the concept more clear, here is an example.

## circularity

**Type: ****float**

**Example:** 0.58

Measure for the similarity of the object shape to a circular shape. It is calculated by the following formula:

, where A and L are the area inside and the length of the object's perimeter. The downside of this approach is that due to pixel quantization, the value for circularity can even reach values larger than 1, which makes these values harder to interpret.

## circularity_ISO

**Type: ****float**** [0-1]**

**Example:** 0.89

Measure for the similarity of the object shape to a circular shape as defined by ISO standards

## solidity

**Type: ****float**** [0-1]**

**Example:** 0.24

The solidity parameter describes the ratio between the area of the object and the area of its convex hull. The latter is the smallest convex shape enclosing the object.

Solidity tells us how a "filled-in" or "complete" object appears in an image. Solidity is a score that helps us understand if the shape of the detected objects is more like a complete, solid object or if more open or irregular.

## eccentricity

**Type: ****float**** [0-1]**

**Example:** 0.74

The eccentricity measures the ratio of the major half-axis to the minor half-axis of the ellipse best describing the size and shape of the object. (See Major-/Minor axis). The exact formula is given by

A circle has an eccentricity of zero, the more oblong the shape the larger the eccentricity gets.

## equivalent_diameter

**Type: ****float**

**Example:** 25.7

The diameter of the circle which possesses the same area as the object.

## extent

**Type: ****float**** [0-1]**

**Example:** 0.68

The extent is the ratio of the object area to the area of its bounding box

## minor- / major_axis_length

**Type: ****float**

**Example:** 26.8

The major and minor axis length refers to the axis of the ellipse best describing the size and shape of the object.

In other words, the 2D shape of an object can be thought of as a 2D probability map. Imagine it like a cloud, where the object is more likely to be found in some areas of the cloud than others. This cloud-shaped probability map can often be described using a mathematical shape called an ellipse. The ellipse represents where the object is most likely to be located and is defined by two important measurements: the 'Major Axis,' which is like its longest stretch, and the 'Minor Axis,' which is the shorter stretch, sort of like the object's width and height.

If you have any questions, please send us an email at support@ikosa.ai.

# Related articles

Copyright 2016-2021 KML Vision GmbH. IKOSA® is a registered EU trademark.