Vuzix Corporation - ビュージックス コーポレーション - Vuzix 3D Watermark Detection

1. REVISION HISTORY
2. INTRODUCTION
2-1. SCOPE
2-2. CONVENTIONS
3. FUNCTIONAL DESCRIPTION
3-1. THEORY OF OPERATION
3-1-1. Overview
3-1-2. State Determination
3-1-3. Reliability Measures
3-1-4. Positive Watermark Blanking Feature
3-1-5. Watermark Processing Caveats
3-2. WATERMARK BIT PATTERN DEFINITIONS.
3-3. EXAMPLE WATERMARK IMAGES
Table 1. Watermark Bit Patterns Figure 1 - Watermark Sample Areas
Figure 2. Side By Side Cross View Example

1. Revision History

Revision : 01
by : SKP
Date : 05/04/2009
Changes : Initial Version

2. INTRODUCTION

2-1. Scope

This document describes the definition of the video watermark that can be detected by the Vuzix iWear to automatically control the video settings of the viewer.

2-2. Conventions

The following conventions will be used to indicate the various radices for values:

• Hexadecimal: 0x5A or ‘5A’h
• Binary: ‘01011010’b
• Decimal: 90 or ‘90’d
• Don’t Care: ‘00x’b (3 bit quantity, top 2 bits are 0, bottom bit is a don’t care)

The following conventions are used to indicate the 3D modes:

• Side By Side: The video is compressed horizontally so that the left half of the screen contains data for one eye and the right half contains data for the other eye.
• Field Sequential: The video is compressed so that field 1 contains data for one eye and field 2 contains data for the other eye.
• Anaglyph: The video contains left eye data encoded with one color filter and right eye data encoded with a different color filter. The data for both eyes is contained in the same frame.

The following conventions are used to indicate the 3D view:

• Parallel View:
  • Side By Side – Left Eye Data is presented on the left half of the screen.
  • Field Sequential – Left Eye Data is presented in Field 1.
  • Anaglyph – Left Eye Data is filtered using the first color described.
• Cross View:
  • Side By Side – Right Eye Data is presented on the left half of the screen.
  • Field Sequential – Right Eye Data is presented in Field 1.
  • Anaglyph – Right Eye Data is filtered using the first color described.

3. Functional Description

3-1. Theory of Operation

3-1-1. Overview
The Vuzix iWear can be set into an “Automatic 3D” mode that analyzes each line of video looking for a pre-defined video pattern in the luminance channel. In this mode, the video is processed in real time looking for a specified 8-bit pattern to be present at the bottom of the screen.

3-1-2. State Determination
The watermark data comprises the full width of the video stream for several lines at the bottom of the screen. The frame is divided into 8 equal sections as shown in Figure 1. The sample area for each section ignores the first and last 12% of each section as to remove artifacts from compression in the transition areas. Each section is analyzed and determined to
be 1 of the possible states. A positive ‘1’b is defined as the average luminance for the section being greater than ‘140’d. A positive ‘0’b is defined as the average luminance for the section being less than ‘80’d. The state is undetermined for and average luminance between ‘80’d and ‘140’d. Any section being in an undetermined state causes a non-detected watermark. Any bit pattern that is valid but undefined in Table 1 is determined to be a non-detected watermark. All bit patterns that are currently marked as “RESERVED FOR FUTURE USE” in Table 1 are determined to be a non-detected watermark. As the watermark algorithm is enhanced and further implemented, support will be added for detection of these bit patterns and this document will be updated.


Figure 1 - Watermark Sample Areas

3-1-3. Reliability Measures
In order to maintain a high reliability rate and reduce the occurrence of false positives in this mode, the watermark is required to be repeated for a minimum of 4 consecutive lines and must also be present for a minimum of 4 frames. The watermark data is required to be present for the entire time that the content creator wants the video to be in a non-2D mode. Once no watermark is detected, the iWear will automatically revert to 2D mode. There is also a small amount of hysteresis that is applied when returning from a 3D mode to normal 2D mode. This hysteresis is approximately 400 milliseconds and serves to reduce the probability of falsely detecting a return to 2D mode.

3-1-4. Positive Watermark Blanking Feature
Once the iWear has detected a positive watermark, all lines containing the watermark data are automatically blanked out and replaced with black. This is done by sampling the luminance level of the black data packet and replacing all pixels on the watermarked lines with that shade of black. This helps maintain uniformity of the black levels throughout the image and provides a less distracting experience for the user.

3-1-5. Watermark Processing Caveats
Because the watermark detection is accomplished by processing the video stream after the video decoder, it is susceptible to detection failure based on the particular settings of the user. The valid bit detection thresholds have been chosen to work over a very broad set of user controlled settings as well as maintaining a small enough delta to prevent false positives from occurring.

3-2.　Watermark Bit Pattern Definitions

Table 1 - Watermark Bit Patterns

3-3.　Example Watermark Images

Figure 2 - Side By Side Cross View Example

This document contains information which is confidential and proprietary to Vuzix Corporation and shall not be reproduced, transferred, or disclosed to others for any purpose without prior written permission from Vuzix Corporation.