Gracenote, Inc.

Roku, Inc.

Zeitera, LLC

Peter Wendt

Jose Pio Pereira

Mihailo M Stojancic

Sunil Suresh Kulkarni

Shashank C Merchant

Prashant Ramanathan

Kyle Brocklehurst

Accurately detection of logos in media content on media presentation devices is addressed. Logos and products are detected in media content produced in retail deployments using a camera. Logo recognition uses saliency analysis, segmentation techniques, and stroke analysis to segment likely logo regions. Logo recognition may suitably employ feature extraction, signature representation, and logo matching. These three approaches make use of neural network based classification and optical character recognition (OCR). One method for OCR recognizes individual characters then performs string matching. Another OCR method uses segment level character recognition with N-gram matching. Synthetic image generation for training of a neural net classifier and utilizing transfer learning features of neural networks are employed to support fast addition of new logos for recognition.

Logo recognition in images and videos

Audio distortion compensation methods to improve accuracy and efficiency of audio content identification are described. The method is also applicable to speech recognition. Methods to detect the interference from speakers and sources, and distortion to audio from environment and devices, are discussed. Additional methods to detect distortion to the content after performing search and correlation are illustrated. The causes of actual distortion at each client are measured and registered and learnt to generate rules for determining likely distortion and interference sources. The learnt rules are applied at the client, and likely distortions that are detected are compensated or heavily distorted sections are ignored at audio level or signature and feature level based on compute resources available. Further methods to subtract the likely distortions in the query at both audio level and after processing at signature and feature level are described.

Robust audio identification with interference cancellation

A mobile device responds in real time to media content presented on a media device, such as a television. The mobile device captures temporal fragments of audio-video content on its microphone, camera, or both and generates corresponding audio-video query fingerprints. The query fingerprints are transmitted to a search server located remotely or used with a search function on the mobile device for content search and identification. Audio features are extracted and audio signal global onset detection is used for input audio frame alignment. Additional audio feature signatures are generated from local audio frame onsets, audio frame frequency domain entropy, and maximum change in the spectral coefficients. Video frames are analyzed to find a television screen in the frames, and a detected active television quadrilateral is used to generate video fingerprints to be combined with audio fingerprints for more reliable content identification.

Media content identification on mobile devices

The overall architecture and details of a scalable video fingerprinting and identification system that is robust with respect to many classes of video distortions is described. In this system, a fingerprint for a piece of multimedia content is composed of a number of compact signatures, along with traversal hash signatures and associated metadata. Numerical descriptors are generated for features found in a multimedia clip, signatures are generated from these descriptors, and a reference signature database is constructed from these signatures. Query signatures are also generated for a query multimedia clip. These query signatures are searched against the reference database using a fast similarity search procedure, to produce a candidate list of matching signatures. This candidate list is further analyzed to find the most likely reference matches. Signature correlation is performed between the likely reference matches and the query clip to improve detection accuracy.

Media fingerprinting and identification system

Methods, apparatus, systems and articles of manufacture of logo recognition in images and videos are disclosed. An example method to detect a specific brand in images and video streams comprises accepting luminance images at a scale in an x direction Sx and a different scale in a y direction Sy in a neural network, and training the neural network with a set of training images for detected features associated with a specific brand.

Audio distortion compensation methods to improve accuracy and efficiency of audio content identification are described. The method is also applicable to speech recognition. Methods to detect the interference from speakers and sources, and distortion to audio from environment and devices are discussed. Additional methods to detect distortion to the content after performing search and correlation are illustrated. The causes of actual distortion at each client are measured and registered and learnt to generate rules for determining likely distortion and interference sources. The learnt rules are applied at the client, and likely distortions that are detected are compensated or heavily distorted sections are ignored at audio level or signature and feature level based on compute resources available. Further methods to subtract the likely distortions in the query at both audio level and after processing at signature and feature level are described.

Content identification methods for consumer devices determine robust audio fingerprints that are resilient to audio distortions. One method generates signatures representing audio content based on a constant Q-factor transform (CQT). A 2D spectral representation of a 1D audio signal facilitates generation of region based signatures within frequency octaves and across the entire 2D signal representation. Also, points of interest are detected within the 2D audio signal representation and interest regions are determined around selected points of interest. Another method generates audio descriptors using an accumulating filter function on bands of the audio spectrum and generates audio transform coefficients. A response of each spectral band is computed and transform coefficients are determined by filtering, by accumulating derivatives with different lags, and computing second order derivatives. Additionally, time and frequency based onset detection determines audio descriptors at events and enhances descriptors with information related to an event.

Audio content fingerprinting based on two-dimensional constant Q-factor transform representation and robust audio identification for time-aligned applications

Video sequence processing is described with various filtering rules applied to extract dominant features for content based video sequence identification. Active regions are determined in video frames of a video sequence. Video frames are selected in response to temporal statistical characteristics of the determined active regions. A two pass analysis is used to detect a set of initial interest points and interest regions in the selected video frames to reduce the effective area of images that are refined by complex filters that provide accurate region characterizations resistant to image distortion for identification of the video frames in the video sequence. Extracted features and descriptors are robust with respect to image scaling, aspect ratio change, rotation, camera viewpoint change, illumination and contrast change, video compression/decompression artifacts and noise. Compact, representative signatures are generated for video sequences to provide effective query video matching and retrieval in a large video database.

Growing community of inventors

San Jose, CA, United States of America

Peter Wendt