Image and video analysis. Image classification and object recognition

Today we are publishing the seventh lecture from the course “Image and Video Analysis” given by Natalia Vasilyeva at the Computer Science Center in St. Petersburg, which was created on the joint initiative of Yandex Data Analysis School, JetBrains and CS-club.



In total, the program has nine lectures, of which have already been published:

1. Introduction to the course "Image and video analysis" ;
2. Basics of spatial and frequency image processing ;
3. Morphological image processing ;
4. Building the signs and comparing images: global signs ;
5. Construction of signs and comparison of images: local signs ;
6. Similarity search. Search for fuzzy duplicates .

Under the cut you will find a plan of the new lecture and slides.


')
Why compare images:

• What to consider as similar objects.
• What is a recognized object.

How difficult is the task of selecting an object:

• Difficulties: different angles / poses of the object.
• Difficulties: different scale.
• Challenges: changing lighting.
• Difficulties: background.
• Difficulties: overlap.
• Difficulties: deformable object.
• Difficulties: intraspecific differences.

How to select common properties of objects from one category:

• General solution scheme.
• What signs to use.
• How to train a classifier.
• Various options for marking the training set.
• Models: Generative vs. Discriminative.
• Discriminative methods.

Classification for different subtasks and object types:

• Definition of the category of the object.
• Category definition: example.
• Select object.
• Using a classifier.
• Use a sliding window.
• Add information about the spatial location in the model "bag of words".
• You can use the pyramid.
• Jittering

Violo-Jones detector:

• Part arrangement models.