Menu Close

Image classification

Image classification is process in computer vision that can classify an image based on its visual content is referred to as mage classification. While humans find it easy to detect objects, robust image classification remains a challenge in computer vision applications. The process of assigning land cover classes to pixels is known as image classification. Water, urban, forest, agriculture, and grassland are just a few examples of classes. The most important aspect of digital image analysis is image classification.

The task of extracting information classes from a multi-band raster image is referred to as image classification in GIS. The image classification raster can be used to create thematic maps. There are two types of classification based on the interaction between the analyst and the computer during classification: supervised and unsupervised.

Also it is the process of categorizing pixels in an image into categories or classes of interest. Examples include built-up areas, bodies of water, green vegetation, bare soil, rocky areas, clouds, and shadows. The relationship between the data and the classes into which they are classified must be well understood in order to classify a set of data into different classes or categories.

what to do classify image ?

To accomplish this with a computer, the computer must first be I trained. well defined Training sample is very crucial , second, Classification techniques should originally developed, and thirdly, Pattern Recognition research was conducted.


Learning algorithms are divided into two types: supervised and unsupervised learning techniques.
The distinction is based on how the learner provide data.

what is the objective of image classification ?

The goal of image classification is to identify and depict the features in an image as a distinct gray level in terms of the object or type of land cover these features actually represent on the ground.

and also the classification process’s goal is to group all pixels in a digital image into one of several land cover classes, or “themes.” This classified data can then be used to create thematic maps of the land cover depicted in an image. Typically, multi-spectral data are used for classification, and the spectral pattern present in the data for each pixel is used as the numerical basis for categorization.

Spectral reflectance curve for Image classification

Supervised Classification and Unsupervised Classification are the two main classification methods in GIS. Unsupervised (calculated by software) and supervised (human-guided) classification are the two major types of image classification techniques.

  1. Supervised Classification

To classify an image, supervised classification employs the spectral signatures obtained from training samples. You can easily create training samples to represent the classes you want to extract using the Image Classification toolbar. You can also easily create a signature file from the training samples, which is then used to classify the image by the multivariate classification tools.

The classifier has the benefit of an analyst or domain knowledge, which can be used to guide the classifier in learning the relationship between the data and the classes. Using this prior knowledge, the number of classes and prototype pixels for each class can be identified. When prior knowledge is available for some classes but not for others, or when some dates are available but not others in a multi-temporal datasets, a combination of supervised and unsupervised methods can be used for partially supervised image classification.

  1. Training sites (also known as testing sets or input classes) are chosen based on the user’s knowledge.
  2. The user also specifies how similar other pixels must be in order to be grouped together.
  3. These bounds are frequently set based on the spectral characteristics of the training area, plus or minus a certain increment often based on “brightness” or reflection strength in specific spectral bands.
  4. based on the idea that a user can choose sample pixels in an image that represent specific classes and then instruct image processing software to use these training sites as references for the classification of all other pixels in the image.
  5. The user also specifies the number of classes into which the image is classified.

Supervised Classification Procedures

A training sample, or a collection of data points known to have come from the class of interest, is required by a supervised classification algorithm for each class. The classification is thus based on how close each training sample is to each point to be classified. We won’t try to define “close” other than to say that both geometric and statistical distance measures are used in practical pattern recognition algorithms.

The training samples are representative of the analyst’s known classes of interest. Supervised classification methods are those that rely on the use of training patterns. In supervised classification, representative samples are chosen for each land cover class. The software then applies these “training sites” to the entire image. A typical supervised classification procedure consists of three basic steps, which are as follows:

The framework of supervised classification

  1. Training stage: The analyst identifies representative training areas and creates numerical descriptions of each land cover type of interest in the scene.
  2. Generate signature file: Each pixel in the image data set is classified according to the land cover class that it most closely resembles. If a pixel is not sufficiently similar to any training data set, it is typically labeled ‘Unknown.’
  3. The output stage, also known as Classify: The outcomes can be used in a variety of ways. Thematic maps, tables, and digital data files are three common types of output products that become input data for GIS.
Image classification supervised

2. Unsupervised classification

The outcomes (groupings of pixels with common characteristics) of unsupervised classification are based on software analysis of an image without the user providing sample classes. The computer employs techniques to determine which pixels are related and classifies them. The user can specify which algorithm the software will use and the desired number of output classes, but the software does not help with classification.

The steps for running an unsupervised classification

  1. Generate clusters: on this step, the software clusters pixels into a set number of classes. So, the first step is to assign the number of classes you want it to generate. In addition, you have to identify which bands you want it to use.
  2. Cluster pixel into spectral classes
  3. label cluster to corresponding to information classes
  4. Evaluate the result

We like your comment