研究目的
To propose a novel illumination correction algorithm and robust feature extraction method for dermoscopy images to improve the identification of melanoma.
研究成果
The proposed WGWS algorithm for illumination correction combined with Gabor and LMeP feature extraction improves melanoma detection accuracy, outperforming other methods. It effectively handles uneven illumination and extracts relevant differential structures, but future work should address rotation and scale invariance.
研究不足
The image descriptor is not invariant to rotation and scale changes in the images.
1：Experimental Design and Method Selection:
The methodology involves illumination correction using a proposed Weighted Gaussian Window Scanning (WGWS) algorithm in the CIELUV color space, followed by texture feature extraction using Gabor filters and Local Mesh Patterns (LMeP). This is designed to handle uneven illumination and extract physiologically relevant differential structures for melanoma detection.
2：Sample Selection and Data Sources:
A publicly available dataset of 200 dermoscopy images (160 nevus, 40 melanoma) acquired at Hospital Pedro Hispano, Matosinhos, with images at 20x magnification and 765x573 resolution, manually segmented and classified by a dermatologist.
3：List of Experimental Equipment and Materials:
Dermoscope for image acquisition, computer with software for processing (e.g., Weka for classification).
4：Experimental Procedures and Operational Workflow:
Convert images to CIELUV space, apply WGWS for illumination correction on the L channel, extract texture features using Gabor filters and LMeP, classify images using 1-NN and SVM classifiers with 10-fold cross-validation.
5：Data Analysis Methods:
Performance evaluation using classification accuracy and area under ROC curve, comparison with homogeneous texture features and homomorphic filtering.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
