releasenotes3_0.htm

A tutorial and open source code for finding edges and corners based on the filters used in primary v

<FONT SIZE=2><P>20</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>12</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>8</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>5</FONT></TD>
</TR>
<TR><TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>Filter</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>65%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>59%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>54%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>51%</FONT></TD>
</TR>
<TR><TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>Interpret</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>33%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>38%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>40%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>39%</FONT></TD>
</TR>
<TR><TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>Display</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>36%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>23%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>13%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>7%</FONT></TD>
</TR>
<TR><TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>Write file</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>47%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>48%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>47%</FONT></TD>
<TD WIDTH="20%" VALIGN="TOP">
<FONT SIZE=2><P>46%</FONT></TD>
</TR>
</TABLE>

<FONT SIZE=2><P>Compared to the version 2.1 code, the time required to process features is 45% when filter size is 20 and 81% when filter size is 8.</P>
<P>Fixed inconsistencies in the code about what is light or dark. Made the orientation and coordinate system consistent. The X-axis points right; Y-axis points down; positive rotation is clockwise. An orientation of 0 degrees corresponds to the vector (0,1) with dark in the region where x is negative and light for positive x.</P>
<P>Added lateral antagonism. During the initial pass, features that fall below threshold are retained. Their effective strength is increased or decreased based on neighboring features. </P>
<P>Added corner detection. After finding the main feature, steer to 90 degrees from it and look for another feature. If there is any, it is a 2D interest point (corner, X, T, high curvature, blob, etc.) If not, it is an edge or bar. [To do: this needs more work.]</P>
<P>Removed steering to +/- 45 degrees. These responses could be used to distinguish certain classes of interest points, but this check has never been implemented.</P>
<P>Changed "edges only" behavior. Previously, the code checked for edges and bars. If the "edges only" view was selected, display of bars was suppressed. Now, at all locations only edges are sought. [To do: in this case we don't need to look at large filters and can save time by omitting large filter correlations.]</P>
<P>Implemented test images for corners.</P>
<P>Corrected the answers generated for bar testing. Verified that noise-free artificial images of edges and bars are usually good to subpixel resolution. </P>
</FONT><B><I><FONT FACE="Arial"><P>Known problems:</P>
</B></I></FONT><FONT SIZE=2><P>Images of thin bars are often misidentified as edges with a large position error.</P>
<P>Bar width is unreliable and almost always estimated as too wide.</P>
<P>Corner detection sometimes winds up putting the perpendicular edge at the wrong end of the receptive field.</P>
<P>Graphics for drawing bars is sometimes incorrect.</P>
</FONT><B><I><FONT FACE="Arial"><P>Class structure</P>
</B></I></FONT><FONT SIZE=2><P>The program executes the following transformations:</P>
<P>Image &#9;-&gt; big image region &#9;-&gt; filtered&#9;-&gt; feature</P>
<P>&#9;-&gt; small image region&#9;-&gt; filtered</P>
<P>feature list -&gt; Image of features</P>
<P>The classes involved represent images and their subregions, sampling locations, filters, detected features, and graphical outlines of features.</P>
<P>&nbsp;</P>
<P>In versions 1 and 2, there are three odd filters at 60 degree orientations and four even filters at 45 degree orientations.</P>
<P>The even and odd filters share a horizontal orientation. </P>
<P>In version 3, this has been changed so that even and odd filters share a vertical orientation. This is meant to make it easier to detect stereo disparity. Version 3 uses three odd filters at 60 degree orientations and two even filters at 90 degree orientations</P>
<P>&nbsp;</P>
</FONT><B><I><FONT FACE="Arial"><P>References</P><DIR>
<DIR>

</B></I></FONT><FONT SIZE=2><P ALIGN="JUSTIFY">http://home.earthlink.net/~tylerfolsom/</P>
<P ALIGN="JUSTIFY">T.C. Folsom, <I>Neural Networks Modeling Cortical Cells for Machine Vision</I>, Ph.D. Thesis, University of Washington, Seattle, Washington, 1994.</P>
<P ALIGN="JUSTIFY">T.C. Folsom and R.B Pinter, "Primitive features from steering, quadrature and scale," <I>IEEE Transactions on Pattern Analysis and Machine Intelligence</I>, vol. 20, no. 11, November 1998, pp. 1161-1173</P>
<P ALIGN="JUSTIFY">Freeman, W. T., and E. H. Adelson "The Design and Use of Steerable Filters", IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 13, pp. 891-906, 1991</FONT>.</P>
<FONT SIZE=2><P ALIGN="JUSTIFY">Vision Technology Group, Microsoft Research , <I>The Microsoft Vision SDK</I>, Version 1.0, March 1998, </FONT>http://www.research.microsoft.com/research/vision/</P></DIR>
</DIR>
</BODY>
</HTML>