ASME Computers and Information
in Engineering Conference
An ASME paper should be: Clear, concise, complete, and original; with assumptions plainly identified; data and computation results presented with their uncertainty, precise logic, relevance to practice described, and with actual accomplishments of the work plainly stated and honestly appraised.
Paper No.: 
CIE200484

TechArea
Chair: 
Richard Crawford


Title: 
Local Feature Extraction Using ScaleSpace Decomposition


Author(s): 
Dmitriy Bespalov
, Ali Shokoufandeh William C. Regli


Reviewer: 
Only for TechArea & Program Chairs

Affiliation 


Date assigned: 
3/30/04

Return
review by*: 
4/13/04

Return
email: 
rhc@mail.utexas.edu


*Please
inform the TechArea Chair immediately if you cannot complete the review by
the indicated date.
Please
delete the manuscript file from your computer after the review has been
returned to the TechArea Chair.
Place
a check in the boxes that, in your opinion, best describe the following
features of the manuscript.
PAPER PROFILE

Poor 
Marginal 
Acceptable 
Good 
Honor 
Originality
of work 


X



Engineering
relevance 



X


Significance
of contributions 


X



Fit
with ASMECIE scope 



X


Completeness
of the reported work 


X



Acknowledgment
of the work of others 

X




Organization
of the work 



X


Clarity
in writing, Tables, graphs, and illustrations 

X





Yes 
No 
N/A 
In
your opinion, is the technical treatment plausible and free of technical
errors? 
X



Have
you checked the equations? 
X



Are
you aware of prior publication or presentation of this work? 

X


Is
the work free of commercialism? 
X



Is
the paper too long? 

X


YOUR RECOMMENDATION
This
paper is: (check only in one place  justify decision by remarks on next page)
X

Acceptable with Minor Revisions

REMARKS
To
assist the author in revising the paper please separate your remarks into two
clearly identified sections:
(1) Changes which must be made
before publication.
(2)
Those suggestions which, in your opinion, would improve the quality of the
paper but are not essential for publication
Remarks
that are not clearly identified will be assumed to fall into the second
category (not essential).
CHANGES
WHICH MUST BE MADE BEFORE PUBLICATION
The paper adresses the problem of feature extraction on fully or partially scanned 3D objects. It applies an approach which computes the maximal angle along the shortest path between pairs of faces of a 3D triangulated model. Having the these values calculated for each pair of faces, a matrix is created, which is used to extract features by a scalespace decomposition. The proposed approach seems to solve this problem, but the presentation needs further improvements. First of all, the related work section is far too brief. It lacks any goals, and it is more like a list of others work, than a consice review. Please reflect on others contribution, and suggest how scalespace decomposition provides further improvents. In section 3 the mathematical expressions needs more explanation in some places, otherwise they will be incomprehensible for the readers: 1) denotations of the first three paperagraphs are weak. It uses two different denotations for features, and they do not reflect the scale of decomposition. This means that they can overlap if they are on different levels. 2) the third sentence in section 3.1 needs a reference to a publication. 3) the first two equations are not numbered. 4) explain why to use the maximum angle between two adjacent planar surfaces. 5) I beleive that matrix [D(ti, tj)]n*n is rather antisymmetric matrix than symmetric, since the diagonal values a zero. 6) distance vectors vi came out of the blue, what are they? 7) what is sigma and sigma'? 8) numbering and interpretation of the last equation is missing. Figure 2 presents the hierarchical feature decomposition of a part. on the last but one level, I think there is an error. Parallel surfaces that are not connected considered to be one feature, and at the same time the connecting facets are excluded. How can this be? Figure 3: it would be nice to see which node in Figure 3a corresponds to which picture in Figure 3b. Figure 46 present a lot of example cases without proper explanation. It would be better to see just the most representative ones, but in the form of a decomposition tree. 
SUGGESTIONS
FOR IMPROVING PAPER BUT NOT ESSENTIAL FOR PUBLICATION

SUGGESTED
DISCUSSERS
