COMPARISON AND CO-REGISTRATION OF DEMS GENERATED FROM HiRISE AND CTX IMAGES

Images from two sensors, the High-Resolution Imaging Science Experiment (HiRISE) and the Context Camera (CTX), both onboard the Mars Reconnaissance Orbiter (MRO), were used to generate high-quality DEMs (Digital Elevation Models) of the Martian surface. However, there were discrepancies between the DEMs generated from the images acquired by these two sensors due to various reasons, such as variations in boresight alignment between the two sensors during the flight in the complex environment. This paper presents a systematic investigation of the discrepancies between the DEMs generated from the HiRISE and CTX images. A combined adjustment algorithm is presented for the co-registration of HiRISE and CTX DEMs. Experimental analysis was carried out using the HiRISE and CTX images collected at the Mars Rover landing site and several other typical regions. The results indicated that there were systematic offsets between the HiRISE and CTX DEMs in the longitude and latitude directions. However, the offset in the altitude was less obvious. After combined adjustment, the offsets were eliminated and the HiRISE and CTX DEMs were co-registered to each other. The presented research is of significance for the synergistic use of HiRISE and CTX images for precision Mars topographic mapping. * Corresponding author Tel.: +852 2766 4335; fax: +852 233


INTRODUCTION
CTX and HiRISE are two imaging systems on board NASA's MRO spacecraft.They have been widely used to derive Mars topographic products such as DEMs (Digital Elevation Models).Due to the complex environment during flight in space, the boresight parameters of the CTX and HiRISE cameras may change.These changes may lead to an inaccurate exterior orientation (EO) of the instruments and further result in inconsistencies between the 3D information obtained from HiRISE and CTX images.
Inconsistencies between the DEMs derived from the HiRISE and CTX images have been noted in the past.Kim and Muller (2009) showed that the DEMs from the CTX and HiRISE images had offsets of about 30 m.They proposed a flexible coregistration method to match the DEMs generated from HiRISE, CTX and HRSC images.Tao et al. (2014) further presented the inconsistencies among the DEMs generated from HiRISE, CTX and HRSC images.However, the inconsistency problem has not been systematically investigated in the past.Effective co-registration of the DEMs through directly adjusting the image orientation parameters has not been adequately studied.
This paper presents a systematic study of the inconsistencies between the HiRISE and CTX DEMs, and a co-registration method to eliminate the inconsistencies by improving the EO parameters of the images.Section 2 describes the specifications of the HiRISE and CTX camera, and the experimental datasets used in this research.Section 3 presents the discrepancies between the DEMs generated from HiRISE and CTX images.
Section 4 describes the co-registration method in detail.Section 5 reports the experimental analysis results.Finally, the findings are discussed and concluding remarks are presented in Section 6.

DATASET DESCRIPTION
The HiRISE camera features a large imaging sensor with 14 CCDs.Its effective focal length is 12 m.The camera is able to collect Martian surface images with a spatial resolution of about 0.25 m/pixel (McEwen et al., 2007).The CTX has a 350 mm focal length telescope with a 5.7° field of view.The CTX imagery has a spatial resolution of about 6 m/pixel (Malin et al., 2007).
The datasets used in this research include three sets of stereo HiRISE and CTX data covering different regions of Mars.The CTX imagery has a larger coverage that is partly covered by the HiRISE imagery.Each set of data consists of two HiRISE stereo images and two CTX stereo images targeting the same objectives.The first dataset (named "Landing-Sites") was collected from the landing site of the Mars Exploration Rover Spirit.The CTX data cover a region of 14.19° to 14.95° S and 175.24° to 175.77° E, while the HiRISE data cover 14.50° to 14.68° S and 175.45° to 175.56° E. There is an overlap of about 5% between the HiRISE and the CTX data.Figure 1 shows the CTX and HiRISE DEM of this area.The HiRISE DEM covers the central part of the CTX DEM, which is marked by a black frame in Figure 1(a).As marked in Figure 1(b), "Husband Hill" is located in the lower right-hand of the region.Spirit climbed this hill and spent the majority of its time there during the exploration mission.The other two areas are called "Possible-Fissure" and "Rayed-Crater.""Possible-Fissure" is an area containing possible fissure sources for floodwaters."Rayed-Crater" is located in the Arcadia Region and contains a large impact crater.The detailed data specifications including image ID, acquisition date, image size, resolution and the range of the HiRISE and CTX images are listed in Table1.CCD alignments, which is not emphasized in this research).

INCONSISTENCIES BETWEEN THE DEMS GENERATED FROM HIRISE AND CTX IMAGES
The processing of the images, including calibration and add-on components, was implemented using ISIS version 3 provided by the United States Geological Survey.The EO parameters of HiRISE and CTX were acquired through interpolating the spacecraft's trajectory and pointing vectors based on the observation time.A self-adaptive triangulation-constrained image matching method (Wu et al., 2011(Wu et al., , 2012;;Zhu et al., 2007) was used to automatically obtain a large number of corresponding feature points (homogeneous points that represent the same texture) from the images.3D coordinates of the matched points were then obtained by photogrammetric intersection using the image EO parameters, from which DEMs were interpolated.The resolutions of the DEMs were 1 m for the HiRISE DEM and 20 m for the CTX DEM.
The comparisons of the CTX and HiRISE DEMs were carried out for the three selected areas as described previously.Corresponding feature points were identified from the CTX and HiRISE DEMs, and the differences between the corresponding points were used to indicate the offsets between the DEMs.Table 2 shows the statistics of the offsets between the HiRISE DEMs and the CTX DEMs.The general trends of the offsets can be observed in Table 2.For example, CTX DEMs always deviate from the HiRISE DEMs along the north-east direction.For the three cases, the offsets in the longitude (x) direction are from 140 m to 230 m, and in the latitude (y) direction from 380 m to 450 m.The offsets in the altitude (z) are not as obvious as those in the other two directions.Figure 2 shows an example of the discrepancies between the HiRISE and CTX DEMs in 3D view, where obvious shifts between the hill peaks can be observed.

Overview of the Approach
A co-registration model was proposed to reduce the inconsistencies between the HiRISE and CTX DEMs by directly improving the EO parameters of the images.The coregistration method integrated HiRISE and CTX data according to a rigid mathematical method.Due to its relatively wider coverage and more stable sensor structure, the CTX data were regarded as reference data while the HiRISE data were regarded as slave data.Tie points were selected in the overlapping regions of the images.The EO parameters of the slave images (HiRISE images) were adjusted iteratively so that the 3D coordinates of the tie points derived from the HiRISE and CTX imagery could be matched to each other.The framework of the co-registration approach is shown in Figure 3.

Tie Point Selection
Tie point selection was an important step in this research.Due to the large differences (about 20 times) in the image resolution between the HiRISE and CTX imagery, it was difficult to automatically identify tie points from the imagery.Therefore, the tie points were marked manually.They normally represent distinctive features.Figure 4 illustrates the choosing of tie points.In Figure 4, the left side is the HiRISE stereo pair (PSP_001513_1655 and PSP_001777_1650) and the right side is the CTX stereo pair (P01_001513_1654 and P02_001777_1653).

EXPERIMENTAL RESULTS
Three datasets were used in the following experimental analysis: the HiRISE and CTX images collected at Landing-Sites, Possible-Fissure and Rayed-Crater.
After the co-registration process, the new EO parameters of the HiRISE images were used to generate the co-registered DEM.
The comparison results between the CTX DEMs and the coregistered HiRISE DEMs are summarized in Table 3.This table shows that the co-registered HiRISE DEMs were closely aligned with the CTX DEMs.The discrepancies in the longitude and latitude directions were almost removed and there were about 10 m offsets in altitude after co-registration.Taking into consideration that the resolution of the CTX DEM is 20 m and the resolution of the HiRISE DEM is 1 m, the approximately 10 m offset in altitude could have arisen from the DEM resolution differences.Compared with the offsets of several hundred meters before co-registration, the coregistration method was proven to be effective.Figure 5 shows the 3D view of the co-registered DEMs, which shows that the offset between the hill peaks was removed.
To further evaluate the performances of the co-registration algorithm, three reference lines were selected for profile analysis in the dataset of Landing-Sites.As shown in the left side of Figure 6, two east-west lines and one north-south line were selected around Husband Hill, where the DEM had the largest terrain fluctuation.Three profiles were derived for each reference line for comparison purpose, including two profiles interpolated from the original HiRISE DEM and the coregistered HiRISE DEM respectively, and one profile from the CTX DEM. Figure 6 shows that the HiRISE DEM had an obvious discrepancy when compared to CTX DEM before coregistration, while after applying the co-registration method, the differences were reduced significantly for all the three reference lines.
Detailed statistics of the three profiles derived from the CTX and HiRISE DEMs are listed in Table 4.According to the table, the discrepancies were effectively removed after the coregistration.4. Statistics of the elevation differences between the profiles derived from the CTX and HiRISE DEMs.

CONCLUSIONS AND DISCUSSION
In this paper, a co-registration method is proposed to integrate the HiRISE and CTX DEMs.The performance of the developed approach was evaluated using HiRISE and CTX imagery in the Landing-Sites, Possible-Fissure and Rayed-Crater areas.The theoretical analysis and experimental validation showed that the co-registration method was able to co-register the HiRISE and CTX DEMs effectively.Before co-registration, the offsets between the CTX DEM and HiRISE DEM were about 140 m to 230 m in the longitude direction and about 380 m to 450 m in the latitude direction.After co-registration, the offsets were removed and the height discrepancies were also drastically reduced to the meter level.
This research permits the full comparative and synergistic use of HiRISE and CTX images for precision Mars topographic mapping.

Figure 1 .
Figure 1.DEMs of the landing site of the Mars Exploration Rover Spirit.(a) DEM generated from CTX stereo pair, and (b) DEM generated from HiRISE stereo pair (it should be noted that the strip inconsistencies in the HiRISE DEM are due to the un-calibrated CCD alignments, which is not emphasized in this research).

Figure 2 .
Figure 2. Example of the discrepancy between HiRISE and CTX DEMs in 3D view

Figure 3 .
Figure 3. Framework of the co-registration approach

Figure 5 .
Figure 5.Comparison of HiRISE and CTX DEMs in 3D view.Left shows the 3D view of the DEMs before co-registration and right shows the 3D view after co-registration.

Figure 6 .
Figure 6.Profile comparison between the CTX and HiRISE DEMs before and after co-registration.

Table 1 .
Specifications of the CTX and HiRISE stereo images used in this research.

Table 3 .
Discrepancies between the CTX DEMs and the HiRISE DEMs after co-registration.