Object recognition is a computer vision technique for identifying objects in images or videos. ... Using object recognition to identify different categories of objects. Object recognition is a key technology behind driverless cars, enabling them to recognize a stop sign or to distinguish a pedestrian from a lamppost.

Size, color, and shape are some commonly used features. system depend on the types of objects to be recognized and the organization of the model database. Using the detected features in the image, the hypothesizer assigns likelihoods to objects present in the scene.

Refer the following link:

https://www.learnopencv.com/image-recognition-and-object-detection-part1/

The main focus in Artificial Intelligence (AI) or Machine Learning (ML) is to obtain maximum possible optimization for any underline task. It is to be noted that, in AI or ML the main tool is approximation for optimal solution. So the question is how to obtain a best approximate model/method/algorithm/conjuncture/ad hoc method etc., for best optimal solution to AI tasks. Here, the first principal is good approximation for Optimization.

Object recognition is a computer vision technique for identifying objects in images or videos. ... Using object recognition to identify different categories of objects. Object recognition is a key technology behind driverless cars, enabling them to recognize a stop sign or to distinguish a pedestrian from a lamppost.

Size, color, and shape are some commonly used features. system depend on the types of objects to be recognized and the organization of the model database. Using the detected features in the image, the hypothesizer assigns likelihoods to objects present in the scene.

Refer the following link:

https://www.learnopencv.com/image-recognition-and-object-detection-part1/

Some first principles.

Our senses are not meant to show us how the world really is. They exist for survival.

Our eyes are not digital cameras. Take with your eyes a 'photograph' of your cluttered desktop. Close your eyes and read from your mental image what items are on it. You will note that the mental representation is symbolic and restricted to the capacity of working memory.

At a moment only about the solid angle of one letter can be seen sharply. Around that spot the picture gets more and more blurred.

Visual objects fulfill a function. Often it is sufficient that this function is fulfilled rather than the image looks natural. For example, you can underline a sentence with a line, but also with lots of triangles. Then the triangles serve as line. Two coins, an eraser, and pen make up a face, if arranged in appropriate distance relations. Convolutional neural nets not required for overall face recognition.

If you can reduce an image and a human is still able to recognize the relevant objects, then the reduced image is good enough for being processed by a machine.

Skeletons of objects can already be sufficient to recognize them

https://www.eurekalert.org/pub_releases/2019-08/ehs-ssk082019.php

https://www.youtube.com/watch?v=vb3jUZ_sOX0

https://www.nature.com/articles/s41598-019-45268-y

To every class of images there is an algorithm that performs well on it but badly on a different set of images.

To every image there is always a worse one.

If you see an elephant in the clouds, tell me where the trunk, ears, eyes, and legs are.

Regards,

Joachim

Joachim Pimiskern thank you very much, those are the things i was looking for. Very interesting interesting principles, especially the whole line of thought about reducing an image and the skeletons. Have you met any literature where you can learn more of those principles, or can refer to?

Kavitha Pk Aparna Sathya Murthy Muhammad Ali sorry, maybe you didn't understand the question. Your descriptions sound rather like the goals of computer vision than the fundamental principles it relies on.

Dear Samusev,

We, who deal with Physics, take those things as first principle which are experienced over years and felt by observations, but can not be deduced.

All other ideas can be explained and deduced from those- the first principles.

Regards

Nityananda Das

Visual attributes of objects to be recognized (color, size, shape e. .g. Non-equiaxity (morphological orientation) ) actually depend on equipment and conditions (say vawelength) of image recording. Thus appearance of broken polar ice in aerophotograph is often the same as image of bainitic structure of steel as observed by optical microscopy. I mean that the only first principle in image recognition is actually no first principles.

At the same time we may limit a set of dissimilar objects ( say considering only various types of non-metallic inclusions in steel). In such a case the quasi- first principle is a UNIQUE combination of detectable visual features for any particular type.

To add, with no limitations (only cats vs only dogs) you can hardly distinguish UNAMBIGUOUSLY between two types of such pets in terms of their contours.

Object detection, unless you give it a special meaning (?), belongs to the semantic sphere. As far as I know, there is no convincing formalization of this domain.

We know for sure that object detection is subjective.

How many objects in the picture ?

A couple of days ago, I uploaded a preprint entitled:

Histogram learning approach to image recognition

to the server of ResearchGate,. You can access to the full-text which includes

several careless mistakes. The paper gives a new idea and compared our method with the deep learning. I like what you said:

"Often great innovation in science/engineering happens when the new idea is not build on top of the current state of the art or commonly accepted technology. "

Object recognition is categorized under 'soft computing' which means that there are no clear cut principles or definitions on which decisions are based. Solutions are experiment based i.e. for a particular problem at hand classification rules are formulated based on experimental results. This is because of the infinite variations of objects in the real world which precludes any hard and fast definition to be made. Even for a simple concept like 'sky' it is hard to find a precise set of characteristics : sky can be white, gray, blue, black, orange, red, with clouds, without clouds, with or w/o stars, moon etc.