{"id":16251,"date":"2026-04-03T15:54:10","date_gmt":"2026-04-03T19:54:10","guid":{"rendered":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/?p=16251"},"modified":"2026-04-03T16:28:23","modified_gmt":"2026-04-03T20:28:23","slug":"enhancing-semantic-segmentation-using-locally-learned-histogram-features","status":"publish","type":"post","link":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/2026\/04\/03\/enhancing-semantic-segmentation-using-locally-learned-histogram-features\/","title":{"rendered":"Enhancing Semantic Segmentation Using Locally Learned Histogram Features"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">Abstract:<\/h2>\n\n\n\n<p>Semantic segmentation is the task of dividing entire images into non-overlapping regions with per-pixel class labels that correspond to a problem\u2019s objects of interest. To do the task well, researchers introduce techniques of extracting object features from the image cues of shape, color, and texture to give models an improved ability to discriminate between classes. Over the decades, researchers have developed both histogram-based and spatial filter-based methods for extracting texture features in segmentation pipelines. Today, however, convolutional filters are the dominant choice. These are best suited for learning ordered spatial relationships in images as opposed to orderless pixel value distributions within textures. Histogram-based features allow us to represent these informative texture features. To leverage the power of histograms in deep architectures, I propose a module that extracts both convolutional and histogram features and embeds these features within the normal deep feature learning pipeline. The proposed module uses a learnable histogram layer in parallel with convolutional feature extraction then recombines their respective outputs through concatenation. Also, there is a lack of research on learnable histogram layers and the impact that initialization can have on such layers\u2019 performance. With a synthetic dataset based on Voronoi diagrams, I provide insights into learning behavior and initialization strategies for a learnable histogram layer. Second, I evaluate the proposed module\u2019s effectiveness on multiple real-world semantic segmentation datasets, which include the French Land cover from Aerospace ImageRy, Multimodal Material Segmentation, MUUFL-Gulfport, and HyperPRI datasets. Lastly, I discuss strengths and weaknesses of the proposed module, which are shown through results on the related datasets, and I conclude with some ideas for future work on embedding histogram feature learning within deep learning pipelines.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Links:<\/h2>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><a href=\"https:\/\/ufdc.ufl.edu\/UFE0062003\/00001\/pdf\"><img loading=\"lazy\" decoding=\"async\" width=\"90\" height=\"90\" src=\"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-content\/uploads\/sites\/759\/2016\/09\/new-internet-open-web-website-tab-external-link-e1482256597352.png\" alt=\"Link to the paper's website\" class=\"wp-image-434\" style=\"aspect-ratio:1.2403696147845913;width:200px\" \/><\/a><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Citation:<\/h2>\n\n\n\n<pre class=\"wp-block-code\"><code>S. J. Chang, \u201cEnhancing Semantic Segmentation Using Locally Learned Histogram Features.\u201d University of Florida, 2025.<\/code><\/pre>\n\n\n\n<pre class=\"wp-block-code\"><code>@book{chang2025enhancing,\ntitle={Enhancing Semantic Segmentation Using Locally Learned Histogram Features},\nauthor={Chang, Spencer Jordan},\nyear={2025},\npublisher={University of Florida}\n}<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"<p>Abstract: Semantic segmentation is the task of dividing entire images into non-overlapping regions with per-pixel class labels that correspond to a problem\u2019s objects of interest. To do the task well, researchers introduce techniques of extracting object features from the image cues of shape, color, and texture to give models an improved ability to discriminate between [&hellip;]<\/p>\n","protected":false},"author":686,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"single-templates\/single-sidebar-none.php","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"featured_post":"off","footnotes":"","_links_to":"","_links_to_target":""},"categories":[13,15],"tags":[],"class_list":["post-16251","post","type-post","status-publish","format-standard","hentry","category-publication","category-thesis"],"acf":[],"_links":{"self":[{"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/posts\/16251","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/users\/686"}],"replies":[{"embeddable":true,"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/comments?post=16251"}],"version-history":[{"count":1,"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/posts\/16251\/revisions"}],"predecessor-version":[{"id":16255,"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/posts\/16251\/revisions\/16255"}],"wp:attachment":[{"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/media?parent=16251"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/categories?post=16251"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/faculty.eng.ufl.edu\/machine-learning\/wp-json\/wp\/v2\/tags?post=16251"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}