Menu
I have been wondering the real difference between segmentation and classification. Classification using the decision tree algorithm has any difference from segmentation?
519
Detection is the process of identification and classification is the categorization of the object based on a previously defined classes or types. While both are based on discernible properties of the object, classification could take arbitrary boundaries based on the problem domain and independent of detection.
Segmentation increase classification accuracy, since it helps extract features in an image. However, it is different from feature selection which has a better impact on classification. It is effective against noise and loss of information.
Semantic segmentation associates every pixel of an image with a class label such as a person, flower, car and so on. It treats multiple objects of the same class as a single entity. In contrast, instance segmentation treats multiple objects of the same class as distinct individual instances.
Semantic segmentation gives a pixel-level classification in an image, i.e. it classifies the pixels into their corresponding classes, whereas object detection classifies the patches of an image into different object classes and creates a bounding box around that object.
Segmentation, as in Image Segmentation means creating parts of an image into segments which are conceptually meaningful or simple for further analysis. Usually we want to locate objects and boundaries in the images. Simplest example is removing background from foreground.
You can relate it with edge detection and countour detection. There are various methods to do this like clustering using K-Means, compressing image to reduce texture, edge detection or markov fields.
Classification is entirely different. In classification, you want to find LABEL of the given data item. The Labels are usually predefined classes or categories - like whether an email is spam or not, or an image contains a human or animal. Decision tree is one of the approaches to do this.
There have been experiments on segmenting images with the help of classification algorithms. Nikamanon's method (link in reference) tries to create too many segments using cut algorithm and use classification to combine them based on whether it is a good segment based on human intervention or not. Their algorithm didn't perform very well but gave a sign that classification can be used in this task.
p.s. Segmentation is more related to clustering than classification.
75
If you want to know the difference between decision trees (used for classification) and segmentation trees (used for segmentation), a brief explanation is:
Decision trees: optimize for purity of leaf nodes (i.e., they want to classify as good as possible
Segmentation trees: optimize for a "good segmentation of the data", not for purity. A "good segmentation" could be that some classification algorithm (for example logistic regression) performs well on the population segments in the leaves. So the tree doesn't do the classification, but rather tries to find segments of the population for which another model works well. The end goal may not be the classification, but rather the segments themselves (i.e. if you want to identify customers that behave a certain way)
34
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With
