Automatically Extract Image Properties

This guide outlines how to configure the extraction of properties from Images on Kolena.

Configuring Image Property Extraction

1. Navigate to Dataset Details

Scroll down to the "Details" page of your dataset.

Navigating to Image Property Configuration

2. Select Image Fields and Properties

Identify and select the image field(s) from your dataset that you want to analyze. Also select the properties of the field(s) you wish to extract.

In the example below we extract properties from the locator field.

Select Specific Properties of Interest For Relevant Fields

3. Edit Property Configuration

To make additional properties visible (or to hide existing properties), the configuration can be edited.

This will add/remove properties. The example below shows how to add the size property to the image in the locator field. The properties shown in purple are the automatically extracted properties.

Example of adding size to the list of extracted properties

Example

Example of a Hydrated Image - Note that the purple property indicates that they are auto-extracted

Available Image Properties

The following properties are available for automatic image property extraction:

Feature Name	Brief Description
Aspect Ratio	Ratio of image width to height
Brightness	Average pixel intensity
Contrast	Standard deviation of pixel intensity
Height	Height of image in pixels
Pixel Entropy	Entropy of color distribution
Sharpness	Edge density from Canny edge detection
Size	Product of image height and width
Symmetry	Level of vertical symmetry in image
Width	Width of image in pixels

Feature Descriptions

Aspect Ratio

Aspect ratio measures the ratio of the image width to its height. It can be useful in scenarios where the image shape or dimensions impact the analysis or model performance.

\[ \text{Aspect Ratio} = \frac{\text{Width}}{\text{Height}} \]

Example

The above example illustrates variation in aspect ratio

Brightness

Brightness measures the average pixel intensity of an image, which can indicate how light or dark the image appears. It can be useful in scenarios where the brightness impact the analysis or model performance.

\[ \text{Brightness} = \frac{\sum \text{Pixel Intensities}}{\text{Number of Pixels} \times \text{255}} \]

Example

The above example illustrates variation in brightness

Contrast

Contrast measures the standard deviation of pixel intensities in an image, indicating the degree of variation between light and dark areas. The standard-deviation is normalized and bounded by a constant 200. This value ranges from 0 - 1 with larger values denoting higher contrast in an image. This can be useful in scenarios where the contrast impacts the analysis or model performance.

\[ \text{Contrast} = \min\left(1, \frac{\sigma_{\text{Pixel Intensities}}}{\text{200}}\right) \]

Example

The above example illustrates variation in contrast

Height

Height measures the height of the image in pixels. This is a straightforward dimension indicating the number of pixel rows in the image. This enables analyzing any behaviors that vary with changing height.

\[ \text{Height} = \text{Number of Pixel Rows} \]

Pixel Entropy

Pixel entropy measures the entropy of the color distribution in an image, providing a measure of the image's complexity or randomness. Higher entropy indicates more complexity. This can allow understanding of model behavior at varying levels of complexity.

\[ \text{Pixel Entropy} = H\left(\frac{c_i}{\sum_{i=1}^{N} c_i}\right) \]

\[ \text{where } c_i \text{ is the count of unique pixel } i, \text{ and } H \text{ denotes the Shannon entropy.} \]

Example

The above example illustrates variation in entropy

Sharpness

Sharpness measures the edge density in an image using the Canny edge detection algorithm. This can indicate how clear or blurred an image is. The value is the proportion of edge pixels to total pixels. This can be useful in identifying any discrepancies in model performance as it pertains to the blurriness or sharpness of an image.

\[ \text{Sharpness} = \frac{\text{Number of Edge Pixels}}{\text{Total Number of Pixels}} \]

Example

Illustration of the difference between images with low sharpness and high sharpness

Size

Size measures the product of the image height and width, giving the total number of pixels in the image.

\[ \text{Size} = \text{Height} \times \text{Width} \]

Symmetry

Symmetry measures the level of horizontal symmetry in an image by comparing the left and right halves. The value ranges from 0 to 1, with 1 indicating perfect symmetry. This can highlight any behavioral differences in data that is more symmetrical in nature.

\[ \text{Symmetry} = 1 - \frac{\text{MSE(Left Half, Mirrored Right Half)}}{\text{15000}} \]

Example

The above example illustrates variation in symmetry

Width

Width measures the width of the image in pixels. This is a straightforward dimension indicating the number of pixel columns in the image. This enables inspecting any behaviors that vary with changing width.

\[ \text{Width} = \text{Number of Pixel Columns} \]