A new texture descriptor for handwritten document writer identification

<p class="p1">Writer identification is a critical task in the realm of pattern classification, aimed at determining the authorship of a manuscript based on labeled handwriting sam- ples. This area has <span class="s1">g</span>arnered considerable attention from researchers and has seen significant ad<span class="s2">v</span>ancements in the last t<span class="s1">w</span>o decade<span class="s1">s</span>, propelled by the inte- gration of n<span class="s1">ov</span>el computer vision and machine learning algorithms. Commonl<span class="s3">y</span>, approaches within this fi<span class="s4">e</span>ld rely on calculating local t<span class="s1">e</span>xture descriptors of im- ages. In this <span class="s1">w</span>ork, we propose a n<span class="s1">ov</span>el local t<span class="s1">e</span>xture descriptor method, termed multi-points local binary patterns (MP-LBP), which is an enhancement of the traditional local binary patterns (LBP) descripto<span class="s5">r</span>. Our approach i<span class="s6">n</span><span class="s2">v</span>ol<span class="s1">v</span>es apply- ing the M<span class="s1">P</span>-LBP descriptor to patches surrounding Harris <span class="s1">ke</span>y points and aggre- <span class="s1">g</span>ating the image descriptors into encoded <span class="s1">v</span>ectors using the <span class="s1">v</span>ector of locally ag- gr<span class="s1">eg</span>ated descriptors (VLAD) encoding method. These <span class="s1">v</span>ectors are subsequently classified by a ball tree classifier to associate the document with the most plau- sible write<span class="s5">r</span>. <span class="s7">T</span>o assess the effica<span class="s1">c</span>y of our descripto<span class="s6">r</span>, we conducted <span class="s2">ev</span>aluations on fi<span class="s8">v</span>e publicly accessible handwritten databases: CVL, CE<span class="s6">R</span>UG-EN, CE<span class="s6">R</span>UG- CH, BFL, and IAM. The results of these tests pr<span class="s1">o</span>vide insights into the perfo<span class="s2">r</span>- mance of the MP-LBP descriptor in the cont<span class="s1">e</span>xt of writer identification.</p><p> </p>