Motion Invariance in Visual Environments

TL;DR

This paper introduces a novel theory of visual learning based on motion invariance, enabling unsupervised extraction of convolutional features from video streams, inspired by natural vision and governed by principles akin to mechanics.

Contribution

It proposes a new theoretical framework for visual feature learning driven by motion invariance, differing from traditional supervised convolutional networks.

Findings

Features learned via motion invariance improve information indexes.

The theory provides a computational scheme for unsupervised feature discovery.

Experimental results demonstrate the effectiveness of motion-invariant features.

Abstract

The puzzle of computer vision might find new challenging solutions when we realize that most successful methods are working at image level, which is remarkably more difficult than processing directly visual streams, just as happens in nature. In this paper, we claim that their processing naturally leads to formulate the motion invariance principle, which enables the construction of a new theory of visual learning based on convolutional features. The theory addresses a number of intriguing questions that arise in natural vision, and offers a well-posed computational scheme for the discovery of convolutional filters over the retina. They are driven by the Euler-Lagrange differential equations derived from the principle of least cognitive action, that parallels laws of mechanics. Unlike traditional convolutional networks, which need massive supervision, the proposed theory offers a truly new scenario in which feature learning takes place by unsupervised processing of video signals. An experimental report of the theory is presented where we show that features extracted under motion invariance yield an improvement that can be assessed by measuring information-based indexes.