OSU CS 559 - Boundary Ownership by Lifting to 2.1D (8 pages)

Previewing pages 1, 2, 3 of 8 page document View the full content.
View Full Document

Boundary Ownership by Lifting to 2.1D



Previewing pages 1, 2, 3 of actual document.

View the full content.
View Full Document
View Full Document

Boundary Ownership by Lifting to 2.1D

46 views


Pages:
8
School:
Oregon State University
Course:
Cs 559 - Selected Topics In Computer Graphics And Vision
Unformatted text preview:

Boundary Ownership by Lifting to 2 1D Ido Leichter and Michael Lindenbaum Technion Israel Institute of Technology Computer Science Department Technion Haifa 32000 Israel idol mic cs technion ac il Abstract This paper addresses the boundary ownership problem also known as the figure ground assignment problem Estimating boundary ownerships is a key step in perceptual organization it allows higher level processing to be applied on non accidental shapes corresponding to figural regions Existing methods for estimating the boundary ownerships for a given set of boundary curves model the probability distribution function PDF of the binary figure ground random variables associated with the curves Instead of modeling this PDF directly the proposed method uses the 2 1D model it models the PDF of the ordinal depths of the image segments enclosed by the curves After this PDF is maximized the boundary ownership of a curve is determined according to the ordinal depths of the two image segments it abuts This method has two advantages first boundary ownership configurations inconsistent with every depth ordering and thus very likely to be incorrect are eliminated from consideration second it allows for the integration of cues related to image segments not necessarily adjacent in addition to those related to the curves The proposed method models the PDF as a conditional random field CRF conditioned on cues related to the curves T junctions and image segments The CRF is formulated using learnt non parametric distributions of the cues The method significantly improves the currently achieved figure ground assignment accuracy with 20 7 fewer errors in the Berkeley Segmentation Dataset 1 Introduction The boundary ownership problem is the problem of automatically deciding which of the two image segments abutting a provided curve along an object s boundary in an image is the figure i e owns the boundary and which is the ground Although the image segment corresponding to the occluding object is usually the figure boundary ownership is generally a perceptual characteristic That is the image region that is perceived as lying in front of the ground regions is considered as figure 11 even if this perception contradicts the 3D layout of the scene Examples of this may be seen in the images in Fig 1 taken from the Berkeley Segmentation Dataset BSDS They were first segmented by a human observer after which two additional human observers attributed each of the boundary curves between the segments to one of its two abutting segments Although the water in images a and b occludes the land and the rocks beneath it both human observers marked the land and the rocks as the owners of the shorelines This may explain the name shoreline rather than say sealine Another example is provided in image c where both human observers marked the ducks as the owners of the contact line between the ducks and the water although the water occludes the duck parts below this contact line A different type of example is the occasional perception of a hole as the figure in cases where the background seen through the hole does not appear in the image around the object containing the hole 11 It has been shown that boundary ownership perception is influenced by many factors including local characteristics of the boundary shape e g convexity 6 and orientation termed lower region 16 its global characteristics e g symmetry size and orientation 6 11 the appearance of the region itself e g contrast 11 and also its surroundings e g parallelism and surroundedness 11 The orientations of the intensity level sets in the close neighborhood of the boundary have been shown to bear information about which of the two boundary sides occludes the continuation of the other The orientations of the intensity level sets are measured relative to that of the boundary itself This information may thus be used for estimating the boundary ownership as well 5 All methods for automatic figure ground assignment in images measure a subset of the above factors These measurements which are usually ad hoc are used to make the boundary ownership decisions In some works these measurements are performed for each boundary pixel by analyzing the boundary curve segment and the image region in its neighborhood The decision for each boundary pixel a b c Figure 1 Cases where the boundary side corresponding to the occluding object was not classified as figure by the human observers See text for details is then made independently For example such an approach was taken in 5 where the orientations of the intensity level sets near the boundary were used for deciding which of the boundary sides occludes the continuation of the other Another example is 1 where the convexity lower region and size cues were locally measured for boundary points and then used to independently estimate the figure ground label at each point In one of the methods in 15 the assignment for each boundary pixel was also made independently according to the likelihood of a set of local shape descriptors that was based on a reference set of learnt shapemes 9 These shapemes are prototypical shapes that implicitly encode cues such as convexity and parallelism The figure ground assignment along a boundary curve between two junctions is constant This suggests that better precision can be obtained by using cues related to the whole curve segment and making a curve based rather than a pixel based figure ground assignment Such an approach was taken for example in 13 and in 15 In fact in 15 the figure ground assignment to the curves was made by averaging the local shapeme on the curve The latter method provided significantly more accurate results than the local model and reinforced the advantage of making curve based assignments Curve junctions and the angles between the curves at the junctions provide additional cues for the figure ground labels of the junction curves Thus incorporating this additional information in the boundary ownership estimates might increase the precision further Since each junction provides a cue for the mutual figure ground labels of all its curves rather than on each curve independently the in tegration of the junction cues into the boundary ownership model causes the curve assignments to be coupled Curve junctions in figure ground labeling were used for example in 13 and 15 Both these works used a conditional random field CRF 8 to model the probability distribution function PDF of the random variables


View Full Document

Access the best Study Guides, Lecture Notes and Practice Exams

Loading Unlocking...
Login

Join to view Boundary Ownership by Lifting to 2.1D and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Boundary Ownership by Lifting to 2.1D and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?