Collect machine learning training data for expression recognition

Abstract

Apparatus, methods, and articles of manufacture for implementing a crowdsourced pipeline for generating training examples for a machine learning expression classifier. Crowdsourcing providers actively generate images with expressions based on cues or goals. A cue or goal could be to mimic an expression or present it in a certain way or "break out" of an existing expression recognizer. Images are collected and scored by the same or different crowdsourced providers, and those that meet a first quality criterion are then reviewed by experts. The reviewed images are then used as positive or negative examples in training a machine learning expression classifier.

Description

[0001] Cross References to Related Applications [0002] This application claims priority to U.S. Provisional Patent Application Serial No. 61 / 762,820, filed February 8, 2013, and entitled "SYSTEM FOR COLLECTING MACHINELEARNING TRAINING DATA FOR FACIAL EXPRESSION RECOGNITION," the entire contents of which are hereby incorporated by reference entry, including the text, drawings, claims, tables and computer program listing appendix (if any) and all other matters of this US Provisional Patent Application as if fully set forth herein. technical field [0003] This document relates generally to apparatus, methods, and articles of manufacture for collecting training examples for visual machine learning classifiers of emotions, affective states, action units, and mental state-like expressions; use of such training examples in training machine learning classifiers ; and machine-learned classifiers for emotions, affective states, action units, and expressions similar to mental states. ...

AI Technical Summary

Problems solved by technology

While it is relatively easy to collect examples of posed smiles from the web, it is difficult to collect examples of real or realistic expressions of a large number of emotions such as fear, contempt, anger, disgust, and others

Images