Hidden Compensation and Shadow Enhancement for True Orthophoto Generation
2. Ortho Rectification with Hidden Detection and Compensation
The
hidden areas are detected by using Z-buffer and index map. Z-buffer is a matrix set composed of two 2-D matrixes on the image plane. One component of the matrix set is used to record an object location for an image point. The other one stores the distance between the camera and the object and is initialized with a very large value. The index map is also a 2-D matrix on the ground plane to identify if the ground point is hidden. Superimposing DBM on DTM, the coordinates of the surface object may be determined. Applying these coordinates to collinearity condition equation, the projected image coordinates are calculated and rounded to the nearest image grid for each groundel. Meanwhile the projection distance is calculated and compared to the one stored in the Z-buffer. If the distance newly calculated is greater, the corresponding cell in index map is flagged as hidden. On opposite, the location recorded previously is first retrieved from the Z-buffer to locate corresponding groundel for flagging hidden. The Z-buffer is then updated with the new values. As an example, the projection distances of the surface objects X and Z shown in figure 2 are dX and dZ respectively. Because dX < dZ, on index map the cell corresponding to Z will be flagged hidden, and the information regarding X will be stored in the Z-buffer. Since the foregoing detection searches hiddens in horizonal sequence, certain hidden pixels around very steep surface elements may be overlooked. The possibility is illustrated as M portion in fig.2. To solve the problem, along near vertical surfaces pseudo groundels are introduced in the next phase. To determine the number of pseudo pixels, height difference and projection length are used. For instance, if the left building in fig.2 has a height of 50m, the projection length is 5 image pixels, thus on the vertical walls of the building, pseudo pixels are placed in a spacing no greater than 10m. Applying again the detecion rules for these pseudo pixels completes the hidden searching. Due to numerical rounding errors, isolated points or lines may occure in the index map. A post moving window processing employs majority rule to refine the index map is necessary.
With hidden detection mechanism, enhancing orthographic rectification process to minimize hidden defect is possible as shown in fig. 1. Index maps flagging groundels hidden from the master and the slave images is constructed. For each groundel nonhidden from the master image, gray value is casted according to the projection geometry from the master image. For those hidden ones, the index map of slave images are consulted to check for the possibility of patching. Filling-in directly for hidden areas the data patched from slave images may causes radiometric discontinuities. A seamless mosaic technique is developed to reduce the discontinuity. The buffer zone extended from the boundary of particular hidden area is constructed using a morphology operator named dilation. Within the buffer zone, gray value is calculated from both the master and the slave image by weighting average. The weighting factors are determinated according to the distance between the processing pixel and boundary of the hidden area.
Figure 2, Hidden detection for buildings
3. Shadow Detection and Enhancement Operations
The technique for shadow detection is similar to that for hidden detection. In place of projection geometry, parallel projection is used. The plane of Z-buffer is built perpendicularly to the direction of the sun. In case of lacking such information, this direction may be estimated from the shadow of tall buildings. As shown in fig.3a, the azimuth angle of the sun may be measured from the direction of the shadow, whereas the zenith angle may be calculated from the ratio of the building height to the shadow length. As shown in fig. 3b, the distance stored in Z-buffer is replaced with the path length between the surface object and the local zero height. For convenience, the path length is recorded in negative quantity. Hence, the searching algorithm used for hidden detection may be re-used for shadow detection.
To level the illumination differences around shadow areas, image enhancement technique should be applied. Around the building areas, characteristics of ground features are often dominated by human activities and are usually quite different from place to place. Hence, local enhancement is preferred. Surrounding the shadow area, a buffer zone is first constructed using dilation operator. If presented, rooftops are further excluded from both the shadow area and the buffer zone. This is to keep the similarity of image contents in the shadow area and those in the buffer zone. The histograms for particular pair of buffer zone and shadow area are calculated accordingly. Considering the formal one as the reference histogram, the gray value transformation table for the shadow area is then constructed with histogram matching method. Thereafter, the shadow enhancement may be accomplished area by area.
Figure 3, (a) Direction of sun and the buffer-zone surrounding shadow area, (b) Shadow detection, Z-Buffer and path length from local zero height to object.
