WO2003049450A2

WO2003049450A2 - Methods for multimedia content repurposing

Info

Publication number: WO2003049450A2
Application number: PCT/IB2002/005091
Authority: WO
Inventors: Radu S. Jasinschi
Original assignee: Koninklijke Philips Electronics N.V.
Priority date: 2001-12-04
Filing date: 2002-12-02
Publication date: 2003-06-12
Also published as: CN1600032A; JP2005512215A; EP1459552A2; WO2003049450A3; AU2002351088A1

Abstract

Comprehensive multimedia content re-purposing relating to wireless communications employs content constructs that are compact representations of the content information. For video content, the constructs are content operators that represent 2D image regions and/or 3D volumetric regions for objects within the sequence and characterized by various visual attributes, and are extracted from the video sequence by segmentation utilizing video processing techniques. The constructs are employed for intra- and inter-modality transformation to accommodate resource constraints of the mobile device.

Description

Methods for multimedia content repurposing

This application claims priority as a continuation-in-part of U.S. Patent Application Serial No. 10/011,883 entitled distributed processing, storage and transmission of multimedia information filed December 4, 2001. the content of the above-identified application(s) is incorporated herein by reference.

The present invention is directed, in general, to multimedia content transcoding and, more specifically, to intra- and inter-modality multimedia content transcoding for use under resource constraints of mobile devices.

Multimedia content may take the form of one of the three distinct modalities of audio, visual, and textual, or any combination thereof. Content "re-purposing" refers generally and theoretically to re-formatting, re-scaling, and/or transcoding content by changing the content representation within a given domain, such as: from video to video, video to still graphic images, or natural pictures to cartoons in the visual domain; from natural to synthetic sound in the audio domain; and from full text to summaries in the textual domain. In addition, content may be re-purposed by changing from one domain to another, such as from video to text or from audio to text.

A primary use of content re-purposing is to enable the processing, storage, transmission and display of multimedia information on mobile (e.g., wireless) devices. Such devices typically have very stringent limitations on processing, storage, transmission/reception and display capabilities. Through content re-purposing, a mobile device user may have constant access to multimedia information with variable quality depending upon the circumstances, and by using the best available multimedia modality. Current content re-purposing implementations include primarily speech-to- text, where spoken sounds are analyzed to transform them into vowels and consonants for translation into text to be employed, for example, in answering or response (dial-in) systems. Summarization, which deals almost exclusively with textual information, is also employed. There is, therefore, a need in the art for improved techniques for content re- purposing directed to more general uses. To address the above-discussed deficiencies of the prior art, it is a primary object of the present invention to provide, for use in wireless communications system, comprehensive multimedia content re-purposing employing content constructs that are compact representations of the content information. For video content, the constructs are content operators that represent 2D image regions and/or 3D volumetric regions for objects within the sequence and characterized by various visual attributes, and are extracted from the video sequence by segmentation utilizing video processing techniques. The constructs are employed for intra- and inter-modality transformation to accommodate resource constraints of the mobile device.

The foregoing has outlined rather broadly the features and technical advantages of the present invention so that those skilled in the art may better understand the detailed description of the invention that follows. Additional features and advantages of the invention will be described hereinafter that form the subject of the claims of the invention. Those skilled in the art will appreciate that they may readily use the conception and the specific embodiment disclosed as a basis for modifying or designing other structures for carrying out the same purposes of the present invention. Those skilled in the art will also realize that such equivalent constructions do not depart from the spirit and scope of the invention in its broadest form.

Before undertaking the detailed description of the invention below, it maybe advantageous to set forth definitions of certain words or phrases used throughout this patent document: the terms "include" and "comprise," as well as derivatives thereof, mean inclusion without limitation; the term "or" is inclusive, meaning and/or; the phrases "associated with" and "associated therewith," as well as derivatives thereof, may mean to include, be included within, interconnect with, contain, be contained within, connect to or with, couple to or with, be communicable with, cooperate with, interleave, juxtapose, be proximate to, be bound to or with, have, have a property of, or the like; and the term "controller" means any device, system or part thereof that controls at least one operation, whether such a device is implemented in hardware, firmware, software or some combination of at least two of the same. It should be noted that the functionality associated with any particular controller may be centralized or distributed, whether locally or remotely. Definitions for certain words and phrases are provided throughout this patent document, and those of ordinary skill in the art will understand that such definitions apply in many, if not most, instances to prior as well as future uses of such defined words and phrases.

For a more complete understanding of the present invention, and the advantages thereof, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, wherein like numbers designate like objects, and in which:

Fig. 1 depicts a data processing system network employing content re- purposing according to one embodiment of the present invention; Figs. 2A through 2C illustrate infra-modality visual content re-purposing according to one embodiment of the present invention; and

Fig. 3 illustrates inter-modality content re-purposing utilizing compact information according to one embodiment of the present invention.

Figs. 1 through 3, discussed below, and the various embodiments used to describe the principles of the present invention in this patent document are by way of illustration only and should not be construed in any way to limit the scope of the invention. Those skilled in the art will understand that the principles of the present invention may be implemented in any suitably arranged device.

Fig. 1 depicts a data processing system network employing content re- purposing according to one embodiment of the present invention. The data processing system network 100 includes a server system 101 and a client system 102. In the example shown, the server 101 and client 102 are wirelessly coupled and interoperable. The server 101 maybe any system, such as a desktop personal computer (PC), a laptop, a "super-computer," or any other system including a central processing unit (CPU), a local memory system, and a set of dedicated chips that perform specific signal processing operations such as convolutions, etc. Data processing system 100 may include any type of wireless communications network, including video, data, voice/audio, or some combination thereof. Mobile (or fixed wirelessly connected) device 102 may be, for example, a telephone, a personal digital assistant (PDA), a computer, a satellite or terrestrial television and/or radio reception system, or a set top box.

Those skilled in the art will recognize that the complete construction and operation of a data processing system network is not depicted in the drawings or described herein. Instead, for simplicity and clarity, only so much of the details of the construction and operation of a data processing system as are either unique to the present invention or necessary for an understanding of the present invention are shown and described. The remainder of the system may be constructed and operate in accordance with conventional practices. Figs. 2 A through 2C illustrate infra-modality visual content re-purposing according to one embodiment of the present invention. In the exemplary embodiment, server 101 is capable of video sequence and/or static image re-purposing for content delivered to client 102.

For video re-purposing, illustrated in Fig. 2 A, a video sequence 201 is fransformed into constructs by construct generator 202. The constructs describe elements of a compact video sequence representation, allowing (a) access to video sequence content information 203, synthesis of the original input video sequence 204 (or the creation of a new video sequence), and (c) compression of the video sequence 205. The constructs are each a compact representation of video content information, with a small number of constructs capable of representing long video sequences.

Construct use goes far beyond just video compression and the like. When transforming a video sequence into a set of constructs, the video sequence is actually redesigned into a new set of building blocks. In video coding, for example, a video sequence is represented by frames or fields in their uncompressed form or by video streams in their compressed form. In this representation, the atomic units are pixels or fields (frames) in the uncompressed form and packages in the compressed form, with the representation being unstructured with respect to video content information.

Video content information is mid-level visual content information given by "objects" such as two dimensional (2D) image regions or three dimensional (3D) volumetric regions characterized by various visual attributes (e.g., color, motion, shape). To generate video content information, the information must be segmented from the video sequence, which requires use of various image processing and/or computer vision techniques. For example, edge/shape segmentation, motion analysis (2D or 3D), or color segmentation may be employed for the segmentation process. Moreover, the compact representation of the segmented video content information is also important. Fig. 2B illustrates segmentation and compaction, in which the input video sequence 201 is processed by segmentation and compaction units 206 and 207 to generate compact video content operators 208. The content operators 208 form part of the video content construct set. Another type of video content constructs is layered mosaics 209, generated by: (i) determining the relative depth information between different mosaics; and (ii) incrementally combining the relative depth information with individual frame from the input source, partial mosaics, and content operators as illustrated in Fig. 2C. The compact video content operators 208 and the layered mosaics 209 in Fig.

2C constitute video constructs which, together with video content segmentation and compaction units 206 and 207, represent the construct generator 202 of Fig. 2A.

In providing an example of construct generation below, the following visual conditions are assumed: the 3D world (scene) is composed of rigid objects; those objects are distributed at different depth levels forming the scene background, which is static (or at least slowly varying) while the foreground comprises a collection of independently moving (rigid) objects; the objects have a local surface which may be approximated as a plane; and the overall scene illumination is uniform.

Given two successive frames I_k__ϊ and I_k taken from a video sequence at instants k — \ and k , respectively, the compact video content operators are generated as follows:

First, register images I_k__x and I_k by comparing the image intensities at each pixel. If I_k__x = ._I(x_Jt_₁,Λ_₁) ^mά = ^JA<Λ> Λ) > ^where (**-ι>JΑ-ι) ∞d fa*,?*) represent the x and y coordinate image pixels at instants k - \ and k , respectively, then images I_k__x and I_k are registered by computing the nine elements of a 3X3 matrix R(-,-) such that:

^4-1 R(2,0) x x_k +R(2,ϊ) y_k +R(2,2) '

_ i?(l,0) x ^ +i?(l,l) x y, + i?(l,2) ^yk" R(2, 0) x x_k + R(2, l) x y_k +R(2,2) ^'

Matrix R(; •) may be calculated in different ways, such as by using the (6 parameter) affine model with R(2, 0) = R(2, 1) = 0 , R(2, 2) = 1 , R(0, 0) = s_x , R(0, \) = r_x , R(l, 0) = r_{y >} and iJ(l,l) = s_y , where s_{x >} s_y , r_x and r_y denote the x and y components of the scaling and rotation (2D) image vectors s and , respectively. Other suitable models include the 8-para- meter perspective model. In any case, the result of registering image I_k_ to image I_k is image / _, .

Next, image velocity is estimated for the registered images I_k__x and I_k , utilizing one of many techniques including energy-based and gradient-based. The resulting image velocity determines the pixel velocity of regions associated with 3D rigid objects moving in a uniform manner, and correspond to the foreground 3D objects and associated 2D image regions.

Based on the results of image velocity estimation and other visual attributes, image regions are then segmented to determine the parts associated with the foreground objects. This results in image regions that may be appropriately post-processed to fill in gaps, with associated Alpha maps.

From the image regions, a compact set of shape templates may be generated via computational geometry techniques. A simple representation is in terms of rectangular shape approximations. For example, mosaics are extended planar images encoding non- redundant information about the video sequence, coming in layers according to the associated relative depth of world regions and generated incrementally through recursive algorithms. At each step of such algorithms, comparison of the last previously- generated mosaic with the current video sequence image generates the new instance of the mosaic. In general terms, the generation of layered mosaics begins with a video sequence {I ,...,I_N} made up of N successive frames each having an associated compact Alpha map a within {a_x,...,a_N} . Each

Alpha map is obtained from the compact video content operator by filing in the interior of mosaic regions, and is a binary image with ones in the interior region and zeros elsewhere. Assuming that information about relative depth—that is, the relative order of each foreground object and also of all of the foreground objects with respect to the background image-has been employed and that the discrimination between each mosaic plane within L levels is possible, a set of L mosaics (Φ¹ ,...,Φ^L} , where the i ^h mosaic Φ' is computed in an initial step r = l by Φ[ = a[I_x and in subsequent steps r = 2,...,N by recursively combining the set of Alpha maps {a₂',..., _N' } with {7,,...,/^} , thereby generating Φ^. for each step r . Finally, a determination is made of any ancillary information complementing the image regions and shape templates and describing image velocity and other visual attributes necessary to fully represent the video content information.

The result of video construct generation is a set of compact video content operators, a set of layered mosaics, and ancillary information. Image re-purposing is directed to reducing the complexity of the images. For example, the image may be transformed into regions of smooth value of color, brightness, texture, motion, etc. One possible general technique for this task is to minimize the cost function: E(I,T) = -I_M(x,y)fώ dy + dxdy + v\r\, (3)

where /(-,^•) denotes the image region R to be estimated, I_M(-,-) denotes the actual (raw) image, and

Actually, the image region R = U,-^ + T , and the overall boundary T encircles the entire region R . The first term in equation (3) determines the "error" between the actual image and the smooth image, the second term determines the "smoothness" term, and the third term is proportional to the boundary length |r| , where v is a constant. For actuation implementation, equation (3) should be appropriately discretized— i.e., approximated by a sum of terms. In analyzing equation (3), it should be noted that /(^■,•) and I_M(;-) denote the visual attribute being smoothed. For example, if smoothing image velocity V(; •) , then

1 (;-) ≡ ¥(; ^■) , etc.

The image may alternatively be transformed into a cartoon image I_c by utilizing a simplified version of equation (3) in which /(-, •) is restricted to piecewise constant values /(•,•) >K . More precisely, for each region R. the value of /(•,•) is approximated as /,■(^■,^■) = K_s , K_t has a constant real value inside region R_t . If μ is a constant, then equation (3) may be approximated by -²E(/, r) = , (5)

y where v = —r . It can be seen that μ

\\l_M(x,y)dxdy K, = mean_Rι (I_M) = . (6)

The cartoonification of /(•, •) creates regions with a constant value for a given attribute. A full cartoonification is accomplished when the region boundaries are marked in black. The cartoon image I_c is a very simplified version of the original image that keeps the main characteristics of the original image I . Visual information transformation from natural to synthetic is one important application of content re-purposing. 3D meshes may be employed for transforming natural 3D objects to synthetic 3D objects; a combination of perspective and projective transformations with 2D meshes may be employed for transforming natural 3D objects to synthetic 2D objects; and 2D meshes and computational geometry tools may be employed for transforming natural 2D objects to synthetic 2D objects. Audio re-purposing includes speech-to-text transformation according to known techniques, with phonemes being generated by speech recognition and then transformed from phonemes to text. In the present invention, the phonemes should be regarded as a compact set of basic elements by which text information is generated utilizing a dictionary as described in further detail below. Inter-modality content re-purposing corresponds to re-purposing multimedia information between different modalities. In general, the framework for inter-modality content re-purposing includes (i) multimedia content segmentation, (ii) template/pattern matching; (iii) use of cross-modality translation dictionaries. When dealing with multimedia information, an overall hierarchy with respect to complexity exists in the three constituent elements (visual, audio and textual) as follows:

Visual (video) — Visual (pictures) -» Audio - Text. (7)

Therefore, transformations across these different modalities should follow the flow defined in equation (7). While not necessarily dictated as a content hierarchy, this patterned is necessitated by the bits required to represent the content within the various modalities. One common technique for re-purposing content according to the flow defined by equation (7) is to transform all visual and audio information into textual description. Video to still image transformation is commonly performed by sub-sampling frames of a video sequence, with transformation of content information with respect to point-of-view (or perspective) being less common. In the present invention, when transforming video to text, a description of the compact video content (video constructs) is given in the textual domain. Similarly, compact image content is transformed to textual description. In video to image transformation, specific regions (information) of the video constructs are accessed by applying compact image content operators to those regions. Fig. 3 illustrates inter-modality content re-purposing utilizing compact information according to one embodiment of the present invention. Generally, content re- purposing across multimedia modalities is performed in the present invention using compact information (e.g., video constructs, image cartoons). Transformation between compact y elements representing a given modality utilizes a compact information format, which is important in transformation from video frames/fields to static frames or text.

Within system 300, separate video, audio and textual inputs 301-303 are employed, with an additional input 304 employed for still images from either an independent input or sub-sampled from video input 301. Compact constructs 305-308 are generated as described above, with inter-modality content re-purposing employing a set of dictionaries (not separately depicted), which translate information between sets of compact content elements in different modalities. Across-modality dictionaries define how the compact content information is described in a given modality, and may be textual and/or based on metadata of a either a proprietary form or employing an agreed standard (e.g., MPEG-7, TV- Anytime, and/or SIMPTE). Translation between elements of different modalities should be performed using these descriptions, which are particularly suited for transforming from video to images. When transforming from video, images or audio to text, the descriptions represent explanations that may be realized at different levels of details. The structure and functionality of dictionaries of this type are described in greater detail in the cross-referenced application identified above and incorporated herein by reference.

The present invention may be implemented on a continuous access content server containing content within a database, to re-purpose content for mobile access of such content. The content may be re-purposed prior to any request for such content by a mobile device (e.g., when the content is loaded for access from the server) or in response to a specific request from a particular device, customizing the content to the resources available within the mobile device. In particular, the present invention may be advantageously employed within wireless communications utilizing Transmission Convergence Protocol (TCP) or Radio Transmission Protocol (RTP) to provide Internet access to customized PDAs, mini-laptops, etc.

It is important to note that while the present invention has been described in the context of a fully functional system, those skilled in the art will appreciate that at least portions of the mechanism of the present invention are capable of being distributed in the form of a machine usable medium containing instructions in a variety of forms, and that the present invention applies equally regardless of the particular type of signal bearing medium utilized to actually carry out the distribution. Examples of machine usable mediums include: nonvolatile, hard-coded type mediums such as read only memories (ROMs) or erasable, electrically programmable read only memories (EEPROMs), recordable type mediums such as floppy disks, hard disk drives and compact disc read only memories (CD-ROMs) or digital versatile discs (DVDs), and transmission type mediums such as digital and analog communication links.

Although the present invention has been described in detail, those skilled in the art will understand that various changes, substitutions, variations, enhancements, nuances, gradations, lesser forms, alterations, revisions, improvements and knock-offs of the invention disclosed herein may be made without departing from the spirit and scope of the invention in its broadest form.

Claims

CLA S:

1. A system 100 for multimedia content re-purposing comprising a controller 200, 300 generating content constructs 208 that are compact representations of content information and include content operators for video content, wherein the content constructs 208 are suitable for transformation of the content within a modality or between modalities.

2. The system 100 according to claim 1, wherein visual content operators 306 for the video content are formed by registering consecutive images within a sequence, estimating image velocity for the registered images, segmenting image regions to identify foreground objects and generating shape templates.

3. The system 100 according to claim 1, wherein the content constructs 208 include layered mosaics 209 for the video content that are derived from Alpha maps for consecutive images recursively combined by cut and paste operations.

4. The system 100 according to claim 1, wherein the content constructs 208 for the video content are employed to transform images within the video content into regions of smooth value for one or more of color, brightness, texture, and motion.

5. The system 100 according to claim 1, wherein the content constructs 208 for the video content are employed to fransform images within the video content into cartoonlike images.

6. A system 100 for multimedia content re-purposing comprising:

- a mobile device 102 capable of selectively accessing multimedia content; and - a server 101 containing multimedia content for transmission to mobile devices, the server 101 comprising a controller 200 generating content constructs 208 that are compact representations of content information and include content operators for video content, wherein the content constructs 208 are suitable for transformation of the content within a modality or between modalities.

7. A method of multimedia content re-purposing comprising generating content constructs 208 that are compact representations of content information and include content operators for video content, wherein the content constructs 208 are suitable for transformation of the content within a modality or between modalities.

8. The method according to claim 7, wherein the step of generating the content operators for video content further comprises:

- registering consecutive images within a sequence; - estimating image velocity for the registered images;

- segmenting image regions to identify foreground objects and generating shape templates.

9. The method according to claim 7, wherein the step of generating the content constructs 208 for video content further comprises recursively combining Alpha maps for consecutive images by cut and paste operations to form layered mosaics 209.

10. The method according to claim 7, further comprising employing the content constructs 208 for the video content to transform images within the video content into regions of smooth value for one or more of color, brightness, texture, and motion.

11. The method according to claim 7, further comprising employing the content constructs 208 for the video content to transform images within the video content into cartoon-like images.

12. A signal comprising multimedia information generated from content constructs 208 that are compact representations of content information and include content operators for video content, wherein the content constructs 208 are suitable for transformation of the content within a modality or between modalities.

13. The signal according to claim 12, wherein visual content operators for the video content are formed by registering consecutive images within a sequence, estimating image velocity for the registered images, segmenting image regions to identify foreground objects and generating shape templates.

14. The signal according to claim 12, wherein the content constructs 208 include layered mosaics for the video content are derived from Alpha maps for consecutive images recursively combined by cut and paste operations.

15. The signal according to claim 12, wherein the content constructs 208 for the video content are employed to transform images within the video content into regions of smooth value for one or more of color, brightness, texture, and motion.

16. The signal according to claim 12, wherein the content constructs 208 for the video content are employed to transform images within the video content into cartoons.