An experimetal study on Content-based Image Classification for Satellite Image Databases
Current art uses metad8E associated with satellite images to facilitate their retrieval from image repositories. Typical metadSD are geographic location, time, and d7 type. Because themetad7E d not indkS1U which regions within an image are obscured bycloudq retrieval with suchmetad7W may prodWk an i...
| Main Authors: | , , , , , |
|---|---|
| Other Authors: | |
| Format: | Text |
| Language: | English |
| Published: |
2001
|
| Subjects: | |
| Online Access: | http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.23.1531 http://cimic.rutgers.edu/~soon/papers/nec01.pdf |
| Summary: | Current art uses metad8E associated with satellite images to facilitate their retrieval from image repositories. Typical metadSD are geographic location, time, and d7 type. Because themetad7E d not indkS1U which regions within an image are obscured bycloudq retrieval with suchmetad7W may prodWk an image within which the region of interest (ROI) for the user is not visible. We report a system that can automaticallydutomati whether an ROI is visible in the image, and can incorporate this into the metadSS for ind0811Wfi images to enhance searching capability. The goal is to annotate each image withmetad0U regard0U a number of ROIs. An experiment with the system annotated 236 AVHRR images of the North Atlantic from a 5-month viewing period with dhWkD788Wfi that expressed the visibility of an ROI centered on Long Island For ground truth, weused the classifications of three human subjects to d18D7EWfi visibility of the same region of interest, and labeled the ROI with the majorityd17DD8W of the three subjects. Partialcloud covermad the humandW87Uq1Wfi88S subjective,and resulted ind isagreements among the subjects. Using randSUk selected 1 training subsets of the images, wefound the two images whose regions were most like those in images for which the LongIsland region was visible. For training subsets, thedW888D7Wfi0 dUqW ed from the two best images prodWkD average Recalland Precision retrieval results jointly in the 75 percent to 80 percent region. Descriptorsdscr ed from those same two images for the test subsets, also prodWk0 average Recalland Precision results that jointly fell in the 75 to 80 percent region. 1 Introducti6 Trad7k8Wfi8 search and retrieval systems for remotely sensed d ata provid facilities to query by sensor characteristics (sensor type, channel or ban. |
|---|