研究目的
To develop a novel technique using deep learning and Neuro-fuzzy classification for efficient image retrieval from large databases, addressing the semantic gap and visual perception issues in content-based image retrieval (CBIR).
研究成果
The proposed CBIR framework, utilizing statistical features and N-F classifier, outperforms other methods with higher accuracy, precision, recall, and F-measure on the Corel-1K dataset. It shows potential for real-time applications but requires further work to reduce computational time and incorporate more parameters or deep learning techniques.
研究不足
The study uses only three features (color, texture, shape) and is limited to the Corel-1K dataset. Computational time and precision could be improved, and the approach may not generalize to larger or different datasets without further optimization.
1：Experimental Design and Method Selection:
The study employs a CBIR framework with feature extraction using statistical features (standard deviation, skewness, kurtosis) from color histograms, followed by classification using Deep Neural Network (DNN) and Neuro-Fuzzy (N-F) classifiers. The methodology includes algorithms for feature extraction and classification.
2：Sample Selection and Data Sources:
The Corel-1K dataset is used, containing 1000 images across 10 categories (e.g., Elephants, People, Mountains).
3：List of Experimental Equipment and Materials:
MATLAB Version R2015 is used for implementation. No specific hardware or other equipment is mentioned.
4：Experimental Procedures and Operational Workflow:
Steps involve extracting RGB values, computing probability histograms, dividing into bins, calculating statistical features, storing features in a database, training classifiers (DNN with 90 neurons and 10 hidden layers, N-F with 20 clusters and step size 25), and testing with query images to retrieve similar images based on classification output.
5：Data Analysis Methods:
Performance metrics include Accuracy, Precision, Recall, and F-measure, calculated using standard formulas. Statistical testing involves standard deviation, skewness, and kurtosis for feature analysis.
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
