6533b873fe1ef96bd12d5fa1

RESEARCH PRODUCT

Bag of words representation and SVM classifier for timber knots detection on color images

Désiré SidibéMohamad Mazen HittaweFabrice Meriaudeau

subject

Computer sciencebusiness.industryFeature extractionComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISIONImage processingPattern recognitionMathematical morphologyThresholdingSupport vector machineComputingMethodologies_PATTERNRECOGNITIONBag-of-words modelHistogramRGB color modelComputer visionArtificial intelligencebusiness

description

Knots as well as their density have a huge impact on the mechanical properties of wood boards. This paper addresses the issue of their automatic detection. An image processing pipeline which associates low level processing (contrast enhancement, thresholding, mathematical morphology) with bag-of-words approach is developed. We propose a SVM classification based on features obtained by SURF descriptors on RGB images, followed by a dictionary created using the bag-of-words approach. Our method was tested on color images from two different datasets with a total number of 640 knots. The mean recall (true positive) rate achieved was (92%) and (97%) for a single dictionary (built only on samples from the first dataset), for the two datasets respectively, illustrating the robustness of our method.

yearjournalcountryeditionlanguage
2015-05-012015 14th IAPR International Conference on Machine Vision Applications (MVA)
https://doi.org/10.1109/mva.2015.7153187