Thursday, February 29, 2024
HomeAIAll about image processing using machine learning|Types, processing methods and application in...

All about image processing using machine learning|Types, processing methods and application in 2022!

machine learning

In recent years, image processing I think that many people want to know what processing methods are available for machine learning image processing and how they are being used.

Here, we will explain an overview of machine learning, image processing methods, and application examples.

Table of Contents

  • Machine learning review
  • What machine learning image processing can do
  • What is Python used in image processing?
  • Four methods used in image processing
    • 1. Pretreatment
    • 2.Inflation
    • 3. Image generation
    • 4. Classification
  • Introduction of technology used in image processing
    • Object identification
    • Object detection
    • Segmentation
  • Image recognition model using machine learning
    • Logistic regression
    • Random forest
    • Boosting
    • SVM (Support Vector Machine)
  • Example of image recognition using machine learning
    • Object Recognition | Inspection of Defective Products in the Manufacturing Industry
    • Face recognition | Entrance/exit management
    • Character recognition|Contract processing
    • Object detection|Forest management
    • Image caption generation
  • Notes on copyright law in image processing
  • Summary

Machine learning review

Machine learning, also known as ML (Machine Learning), is one of the technologies that support AI, and refers to AI acquiring and learning data on its own.

Specifically, it refers to making computers learn in the same way that humans and animals learn naturally through experience. By learning data by machine learning, it becomes possible to recognize the correct data.

For example, by having AI learn a large number of photos of dogs, AI will be able to recognize that the object in the photo is a “dog”.

What machine learning image processing can do

Image processing is the technology by which a computer determines and analyzes what is in an image. Machine learning enables highly accurate image processing.

Image processing can be used for a variety of things, including object recognition, object detection, face recognition, and character recognition.

What is Python used in image processing?

For image classification and object detection, we use a Python library for image processing. So let’s talk about Python.

Python is one of the most popular programming languages ​​because of its simplicity of code.

Image processing requires processing large amounts of data. A program called library is prepared for each process of data analysis.

And Python has many frameworks for processing large amounts of this data.

Four methods used in image processing

We will introduce four methods used in image processing.

  1. Preprocessing
  2. Padding
  3. Image generation
  4. Classification

I will explain each one.

1. Pretreatment

First, we introduce the pretreatment.

Image data preprocessing is to extract meaningful features from an image. We shape and process raw data to create data to be input to machine learning models.

2.Inflation

(4312)

Contrast adjustment

One is contrast adjustment.

Creates an image with enhanced or reduced contrast. To emphasize, set low brightness pixels below a certain level to 0, high brightness pixels above a certain level to 255, and adjust intermediate brightness pixels.

Gamma conversion

The second is gamma conversion.

A transformation used for things such as displaying on a display that changes the luminance value. Luminance value is an index that expresses the degree of brilliance.

Smoothing

The third is smoothing.

Smooth the image as shown in the photo above. There are methods such as applying an averaging filter.

Gaussian noise

The fourth is noise based on Gaussian distribution.

Adds noise to each pixel by adding a generated value based on a Gaussian distribution.

The Gaussian distribution, also known as the normal distribution, is a distribution that is symmetrical about the mean and takes the largest value at the mean heaven.

Salt&Pepper noise

The fifth is Salt & Pepper noise.

It’s like adding salt and pepper noise. Also known as impulse noise.

Reversal

The sixth is inversion.

It literally flips left/right and up/down.

Normalization

The seventh is normalization.

It refers to the process of transforming the range of feature values ​​to fit within a certain range.

Features are characteristics or attributes of data that a model can use to make predictions. To put it more simply, it refers to a variable that serves as a clue for predicting and classifying data.

Rotate

The eighth is rotation.

Rotate the image.

Resize | Scale

The ninth is enlargement and reduction.

Change the size of the image.

3. Image generation

Another method is image generation.

Image generation refers to the technology for generating paintings and automatically processing images and videos. It is also known as a technology that uses deep learning, which is one of the machine learning methods.

Generate using VAE (variational autoencoder), GAN (generative adversarial network), etc.

4. Classification

(Source: https://mavic.ne.jp/ai-aipro/)

Another method of image processing is classification.

Classification is used for image classification, which classifies what is in an image, and object detection, which detects specific objects in an image.

Introduction of technology used in image processing

Many techniques are used for image processing.

  1. Object identification
  2. Object detection
  3. Segmentation

Each is explained below.

Object identification

Object identification, also known as object recognition, refers to the technology of extracting information about objects contained in images.

It is used when verifying whether an object identical to a certain object exists in the image, or when guessing the category of the object in the image.

Object detection

Object detection is a technology for detecting the position of a target object in an image, and although it differs from object recognition, it is often used in conjunction with object recognition because the position of the object is often important when extracting the features of the target object. may occur.

Segmentation

Segmentation is the process of cutting out an area of ​​interest from an image and identifying each subject in the image.

Instead of detecting the whole image or a part of the image, a method of labeling the meaning indicated for each pixel, identifying the area of ​​the object in the image, segmenting the area for each individual and identifying the type of object. There are ways to recognize

A pixel is a unit that represents one small point that makes up a display such as a personal computer.

Image recognition model using machine learning

We introduce an image recognition model that utilizes machine learning.

  1. logistic regression
  2. random forest
  3. boosting
  4. SVM (Support Vector Machine)

I will explain each.

Logistic regression

Logistic regression is one of the simplest and most popular linear class classification algorithms used for supervised learning classification tasks in machine learning.

It is used when you want to predict and analyze the probability that a certain event will occur, and it is said that it is good at distinguishing ambiguous classifications. Classification is done by calculating the probability that the data belongs to each class.

Due to its simplicity, it is often used first in classifiers.

Random forest

Random forest is one of the machine learning models, it is a method of obtaining the result by majority vote of the prediction result of each decision tree using multiple decision trees. It can be used for class classification, regression, clustering, etc.

Decision trees are one of the supervised learning algorithms that can do both regression and classification. Among them, the decision tree sets conditions for the data multiple times and classifies them step by step.

Boosting

Boosting is one of the ensemble learning methods that combine multiple models to improve the prediction accuracy of machine learning.

f:id:St_Hakky:20170728171209j:plain

As shown in the figure above, the models are combined in series and focused on learning the wrong parts. There are various methods of boosting depending on the data weighting method, and typical ones are Adaboost and gradient boosting.

SVM (Support Vector Machine)

A type of machine learning model, it is a very powerful algorithm. Supervised learning can handle classification and regression, but is primarily used for classification tasks.

It has the advantages of good identification accuracy even when the dimension of the data is large and the number of parameters to be optimized is small.

Example of image recognition using machine learning

Machine learning is widely used in various fields. This section introduces an example of image recognition.

  1. Object Recognition | Inspection of Defective Products in the Manufacturing Industry
  2. Face recognition | Entrance/exit management
  3. Character recognition|Contract processing
  4. Object detection|Forest management
  5. image caption generation

I will explain each.

Object Recognition | Inspection of Defective Products in the Manufacturing Industry

First, let’s look at an application example of object recognition.

Some food manufacturing companies, such as Kewpie, are using AI to detect defective raw materials on the production line of food factories.

We are building an algorithm that uses deep learning and image processing technology to learn the rules for distinguishing between good and defective products by learning the video footage of food that flows on the production line of the factory.

Face recognition | Entrance/exit management

Face recognition is also used in various situations, such as entrance/exit management.

Face recognition is one of the biometric authentication technologies, and these technologies are now widely used not only for corporate services but also for banks and national infrastructures where strong security is required.

Character recognition|Contract processing

By using character recognition technology, it is possible to proofread and review packages, advertisements, and various publications, as well as review contracts and application forms. Reducing these tasks will reduce the burden on those who proofread, censor, and review documents.

If the burden is reduced, you will have more time to do the work that you could not do before, and you will be able to work more efficiently.

Object detection|Forest management

Object detection is a technology that detects specific objects from images and videos. This technology is also used in forest management.

It is said to be difficult to automatically generate a land cover classification map and extract changes in land cover by comparing the maps from different periods. A land cover classification map is a classification of whether the land is forest, grassy, ​​or bare. However, the technology of object detection has made it easier to create this map.

Area estimation is also possible, and utilization in monitoring changes in forest areas is also being considered.

Image caption generation

Image caption generation is also one of image recognition using machine learning. In addition to image recognition of the behavior, state, and attributes of each object, it also grasps the relationships between objects appearing in the image, and the scene conditions of the background and location from the image.

Image caption generation is used in applications such as those for visually impaired people who have difficulty recognizing images because it is possible to explain images in natural sentences.

Notes on copyright law in image processing

During image processing, machine learning is performed using images, but at this time it is necessary to pay attention to copyright. There are countless images and photographs that can be freely downloaded on the Internet, but many of them are “copyrighted works.” Therefore, downloading copyrighted material that cannot be said to be for “private use” constitutes copyright infringement.

It is said that copyrighted works can be used in certain cases for the purpose of information analysis such as machine learning, but it is illegal to deviate from private use, so be careful.

Summary

In this article, we have reviewed machine learning, introduced what image processing can do, processing methods, and application examples of image processing.

The basics are explained in an easy-to-understand manner, so if you want to use image processing in your work or research, or if you want to learn more about machine learning, please refer to it.

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Recent Posts

Most Popular

Recent Comments