Content-based Image Retrieval for Asset Management Based on Weighted Feature and K-means Clustering

Assets have different shapes, colors and textures. A shape feature is the main characteristic, which distinguishes each type of asset in addition to the other features, i.e. the color and the texture. The value of features in the image-based asset information retrieval is used as the key field by comparing the similarities between the images. These similarities can be determined based on the differences in the feature values of query images and the images in the database. The closer to zero the difference, the higher the degree of similarities. The degree of similarities will affect the accuracy level of the image retrieval. In this research, image retrieval accuracy and computing time measurement were analyzed. The retrieval accuracy could improve using the property-weighting scheme of weighted features. Moment invariants, color moments and statistical texture are selected to representation shape, color and texture feature. Preprocessing before extraction was done through grayscale, resizing, edge enhancement, histogram equalization. Image asset data clustering was done using K-Means clustering. The research findings suggest that the retrieval accuracy with a total of 400 asset-image data is more than 95% for weighting scheme of Ws (weighted shape) = 50%, Wc (weighted color) = 30%, Wt (weighted texture) = 20% and 10 clusters with the average computing time of 5 millisecond.