Computer vision #2 | Image acquisition, characterization and interpretation

From the moment an image is captured until some information is inferred, various subprocesses are at work. These process the image as a set of pixels to classify and interpret different groups of pixels based on certain criteria. In short, they try to “identify” which objects are contained in a given scene.

Image acquisition, characterization, and interpretation

There are basically six sub-processes in the process of image acquisition, characterization, and interpretation. These in turn can be classified according to their level of abstraction: low, medium, and high level. This set of sub-processes is generally sequential, as the results obtained in one phase usually determine those of the next. The processes involved could easily vary depending on the problem and its complexity.

1. Low-level subprocesses

They cover the area of image processing. They take an image as input and return another one that has been enhanced based on certain parameters or criteria.

1.1. Acquisition

Obtaining the two-dimensional image through a device that captures the three-dimensional world.

1.2. Pre-processing

Image processing to enhance its appearance by improving some of its features.

2. Medium-level subprocesses

Responsible for analyzing the pre-processed image. They receive an image and, after segmenting it into different pixel regions based on characteristics that relate them (proximity, orientation, similarity, etc.) and extracting quantitative information from these regions, they return a classification of the objects according to their category. Their proper functioning is crucial for the success of the overall task.

2.1. Segmentation

Quantitative grouping of the image's pixels into clusters of interest based on their properties.

2.2. Description

Extraction of information based on the relevant features of each object identified in the segmentation.

• External characteristics of the object or region: geometric shape, perimeter, major and minor axes, minimum bounding rectangle, eccentricity, etc.
• Internal characteristics of the object or region: region area, center of gravity, texture patterns (smooth, rough, regular, etc.), color (average, median intensity levels, maximum and minimum intensity values), etc.

2.3. Classification

A particularly dense stage: it recognizes or classifies objects based on their characteristics, grouping them into different categories. This classification is done automatically or, if necessary, with the minimum degree of human intervention possible. It uses various advanced statistical techniques, structural or appearance-based methods, neural networks, genetic algorithms, etc.

3. High-level subprocesses

They understand the scene that has been analyzed in previous stages. The distinction of objects takes context for the overall computer vision process.

3.1. Interpretation

If successful, the machine would have been able to identify the relevant objects for the problem at hand. After this, the range of possibilities is as vast as the number of existing applications, no matter how specific they may be.

4. References

Iván García S., Víctor Caranqui S. (2015). La visión artificial y los campos de aplicación (Universidad Politécnica Estatal del Carchi – Ecuador). Retrieved on October 2021.

This post is also available in Spanish at "Visión artificial #2 | Obtención, caracterización e interpretación de imágenes".