HiRes DECONVOLUTION OF SPITZER INFRARED IMAGES

The algorithm is based on the MCM, reformulated to enable convolutions to be evaluated using the convolution theorem, which enables the use of fast Fourier transforms (FFTs).

Let

• D = acquired image
• P = PSF
• $\tilde{P} =$ reflected PSF, $\tilde{P}(v) = P(-v)$
• fⁿ = current trial image estimate.

The PSF, as used here, represents the response of the telescope optics only, and does not include the spatial response of detector pixels on the focal-plane array. In subsequent discussion we will use the alternative term "point response function" (PRF) to refer to the full spatial response of the measurement system to a point source.

The initial image estimate is customarily set uniformly to unity. Then the single coverage RL algorithm can be described as a series of iterations shown in vectorized form as

$$\begin{equation} f^{n+1} = f^n{{D}\over {f^n*P}}*\tilde{P}, \end{equation} \tag{ 1 }$$

a procedure which consists of three main steps: first, simulate acquisition of the data by convolving the current image with the PSF and create an image of correction factors by dividing the simulated data into the acquired data pixel by pixel. Second, convolve the correction factor image with the reflected PSF. Finally, apply the smoothed correction factors to the current image to produce the next estimate.

Extension to the multiple coverage case involves what is essentially weighted averaging

$$\begin{equation} f^{n+1} = f^n{\frac{\sum _j{{\displaystyle {\bigg [\frac{D_j}{f^n*P_j}\bigg ]}}*\tilde{P}_j}}{\sum _j{U_j*\tilde{P}_j}}}, \end{equation} \tag{ 2 }$$

where U_j is a unit 2D pulse function co-aligned with the jth data pixel.

The extension retains the key attributes of the RL method, such as flux conservation and the non-negativity of the estimated image. Our implementation makes the following two assumptions:

• 1.  
  Isoplanatic PSF. The PSF does not vary over the detector array. The D_j could be decomposed into sub-arrays small enough to assume needed stationarity with a different PSF on each one, but HiRes presently does not handle separate PSFs.
• 2.  
  Constant measurement noise. HiRes assumes that the rms measurement error of each pixel in each observation is constant. Noise weighting each pixel would, however, be feasible, as in Aumann et al. (1990), by dividing each term in the summations by σ²_j.

We have implemented HiRes in a form which is suitable for combining multiple Spitzer basic calibrated data images (BCDs) onto an oversampled grid (typically 1'' for MIPS and 025 for IRAC) and corrects internally for the effect of focal-plane distortion based on the distortion parameters present in the input FITS headers. It is tolerant of missing data (caused, for example, by cosmic ray hits) since the MCM does not require regular sampling of the input images. Since we do not have reliable PSFs we use PRFs (defined as as a cross-correlation of the detector pixel by the PSF) derived from observations. With some computational care the algorithm is modifiable to provide the optional use of PRFs in place of PSFs.

The geometry for the deconvolution is illustrated in Figure 1 showing the overall setting for the calculations, using Equation (2). The trial image is sampled at a level appropriate to the amount of redundant coverage. The data images are generally not aligned with the trial image, and they are not necessarily co-aligned with each other, although for a given Spitzer observation set taken on the same day they would be co-oriented to within one degree of rotation. In this case it is important to note that equal (or nearly equal) $\tilde{P}_j$ can be factored out of partial sums, greatly reducing the number of FFTs that need to be performed. The same is true for the simulation convolutions: resampling to the detector image resolution and division into each member of a co-aligned subset of the D_j can be performed using the trial image convolved once with one of the P_j oriented with the subset.

Zoom In Zoom Out Reset image size

The iterative procedure itself is a relatively simple implementation. HiRes performs various calculations to prepare for it.

• 1.  
  Resampling. Striping is mitigated by correcting accumulations with ratios computed from pixel areas and inclusion counts.
• 2.  
  Edge effects. Abrupt flux transitions at the image boundary can cause high frequency-ringing artifacts. We mitigate this by adopting the included PSF solid-angle weighting technique discussed in Aumann et al. (1990).
• 3.  
  Projection calculation. Some observation scenarios can cover as much as 4 degrees of sky, mandating an exact calculation to perform tangent plane transfers. This is done with 3D homogeneous linear coordinate transformations followed by renormalization to the destination tangent plane.
• 4.  
  Distortion. Spitzer optical distortion is recorded in the form of FITS keywords giving coefficients for quadratic or cubic bivariate Taylor series. These polynomials need to be evaluated for each trial image pixel for each observation image to determine pixel inclusions for resampling. We use the recursive Horner algorithm (published by him in 1819, but known to the Chinese hundreds of years beforehand). It involves successive factorizations of the independent variable to reach a form with nested multiplications (familiar to most students of computer science), which we generalized to the multivariate case with a second recursion on the coefficients, which in this conception are themselves polynomials, with the number of variables reduced by one. The Jacobian of the distortion transformation is needed because pixel areas in Spitzer images can vary by as much as 10%.

For FFTs a suitable convolution pad is added to the trial image. Then dimensions are further increased slightly, up to the nearest higher numbers which are products of powers of the first few primes. A substantial percentage of the execution time is spent evaluating the Fourier transforms, which are RAM and CPU intensive. We use as much memory as we can and use an FFT package that supports threading.

All of the image files used are in the FITS format. The program is invoked with a script run in background mode, and information about its progress is piped to a log file. Various switches set output image registration, resolution, dimensions, and orientation, the latter being important in cases where coverage is long and narrow, and diagonal to the local celestial frame. The trial images and correction factor images from successive iterations can be examined with an FITS viewer during execution. Correction factor images are useful for diagnosis of bad input pixels and poorly-matched PRFs. At convergence they should uniformly be within a percent or so of all 1s. Our largest runs, 4096 × 4096 pixel trial image and 300 256 × 256 input images, take 5 min per iteration on a modern 2 CPU PC with 2 GB of RAM running Linux.

The number of iterations to run is an imposition of a priori knowledge on the part of the user. The literature is extensive on stop rules for RL (see Prasad 2002). In the limit of a large number of iterations, the HiRes output image may be regarded as a maximum likelihood solution to the sky intensity distribution, subject to the constraints of non-negativity and flux conservation. However, the maximization of the likelihood occurs only after convergence is achieved. At each iteration prior to convergence, there is an implicit smoothness bias in the estimated image, and this smoothness decreases as the iterations progress. The optimal termination point for the iterative procedure then occurs when the PRF-convolved estimated image is able to reproduce the observed data to within the error bars of measurement. For the Spitzer data, we have found that this condition is typically achieved after approximately 50 iterations.

The demonstrations shown in Figures 2–4 and the values in Tables 1 and 2 describe the performance of HiRes deconvolution on a bright point source. For this demonstration we use a bright star (at position R.A., decl. (2000): 19:36:57.12 +07:35:23) about 93'' to the NE of the B335 protostar. At MIPS 70 μm this star is too weak and we use the B335 protostar itself to characterize the point-source response in the HiRes image. In Figures 2 and 3 we show the HiRes images and the corresponding mosaic images. We used identical color stretches for both the HiRes and mosaic images to bring out the weakest emission in the diffraction lobes. It may be noted that in the IRAC images at low brightness levels (less than 1% of the peak brightness of the star) there are two point sources within ∼9'' from the star. In the mosaic images they are swamped by the diffraction lobes. HiRes deconvolution achieves sub-arcsec resolution for the IRAC bands. The FWHM values obtained by Gauss fit to the bright star are shown in Table 1 and Figure 4.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 1 shows the typical resolution enhancement in the HiRes images over the corresponding mosaic images. The exact values for the FWHM in a given set of images may differ marginally depending on the number of BCDs (map pixel coverage). However we do not see any significant variations above 10%. In Table 2 we show the level of diffraction residuals observed at Airy rings in some of the HiRes images. To give a conservative estimate we have chosen the images that displayed somewhat high residuals as indicated. A comparison of the mosaic and deconvolved images show that diffraction lobes are removed to a very high degree. In the diffraction residue in the deconvolved images a low-level ringing around bright point sources is seen at levels indicated in Table 2. These are at levels lower than diffraction lobes in the mosaic images by factors of >10. It may be noted that these are conservative estimates obtained using the images as indicated in the table rather than using the best cases in our analysis. Removing the diffraction structures to such low levels more clearly brings out the morphology of the faint emission around bright objects.

We have verified via extensive numerical calculations with both synthetic and real data that the resulting enhancement in angular resolution is approximately three times that obtainable in raw co-added (mosaic) images, and that the fluxes of individual point sources are conserved to a level that is better than 5%. In Figure 5 we show a comparison of the HiRes fluxes with those extracted from mosaic images for a young cluster NGC2316 (Langer et al. 2007).

Zoom In Zoom Out Reset image size

To fully demonstrate the robustness of the deconvolved features we compare the Spitzer HiRes deconvolved images at 8.0 μm with high spatial resolution images observed with Keck at 7.9 μm. The Spitzer and Keck images of a Young Stellar Object (YSO) WL 16 and a multiple star system WL 20 are shown in Figures 6 and 7, respectively. The Keck images in Figures 6 and 7 are reproduced from Ressler & Barsony (2001, 2003). The Keck images were obtained using the JPL's mid-IR camera, MIRLIN (Ressler et al. 1994) with high spatial resolution of 0.3''. The left panels in each figure show the Spitzer 8 μm (a) mosaic, and (b) deconvolved images. The right panels show the Keck images (c) observed with full resolution (0.3''), and (d) smoothed to the resolution of the Spitzer deconvolved images. WL 16 is an embedded intermediate-mass (2–8 M_☉) YSO in ρOph which shows polycyclic aromatic hydrocorbon (PAH) emission features (Waters & Waelkens 1998) and has a unique extended, high-surface-brightness disk which is visible only at mid-IR wavelengths. In the Keck images Ressler & Barsony (2003) observed a disk size of 7'' × 3.5'', corresponding to a disk diameter of ∼900 AU. The disk's major axis is at a position angle of 60°. These features as seen in the Keck image (Figure 6(d)) are very well represented in the deconvolved Spitzer image (Figure 6(b)) demonstrating the fidelity of deconvolution for imaging extended morphology. The Spitzer contour map overlaid on the Keck image (Figure 6(d)) brings out the excellent agreement between the two images. A low intensity blob seen at ∼5'' to the south in the Spizer deconvolved image is not an artefact produced by the deconvolution and its presence in the original data is evident in Figure 6(a).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

WL 20 is classified as an "IR companion system," with WL 20 S exhibiting an embedded protostellar (Class I) spectral-energy distribution (SED) while its two neighbors, WL 20 E and W, each exhibit a T Tauri star (Class II) SED (Ressler & Barsony 2001). The IR companion, WL 20 S, is the dominant luminosity source in the system. The components W20 S and W are blended in the Spitzer mosaic image and they are fully deblended in the HiRes deconvolved image. It may be noted that we used 100 iterations for this deconvolution as it deblended the sources more clearly than using the nominal 50 iterations. The Spitzer deconvolved 8 μm image (Figure 7(b)) is in excellent agreement with the Keck 7.9 μm image (Figure 7(d)). The Spitzer 8 μm flux densities of the components S, E, and W measured using the deconvolved images are 200.6, 153.1, and 47.6 mJy, respectively compared to the Keck 7.9 μm fluxes 123.0, 121.0, and 38.4 mJy, respectively (Ressler & Barsony 2001). The differences in the absolute and relative flux densities may be due to (i) possible differences in the flux calibrations, and (ii) the broader bandwidth by a factor >2 for Spitzer 8 μm than for Keck data, considering the fact the SEDs for components E and W show strong decrease in flux with increasing wavelength while component S shows increasing flux with wavelength (Ressler & Barsony 2001).

The young protostellar objects are so heavily obscured at visual wavelengths that their disks, outflows, and envelopes cannot be observed directly. The Spitzer IRAC and MIPS bands offer a unique opportunity to study protostars, protostellar disks, outflows, and protostellar envelopes and cores. In short-wavelength IRAC bands the outflow cones are observable in scattered light from the protostar through the cavity created by the jets and outflows (Tobin et al. 2007; Velusamy et al. 2007a, 2007b). The emissions in the IRAC bands are also considered to be from the the H₂ rotational lines and possibly some contribution from PAHs (e.g. Noriega-Crespo et al. 2004a, 2004b; Neufeld et al. 2006). Recently Velusamy et al. (2007a) have shown that molecular jets and molecular gas in the bow shocks are observable in the IRAC bands while the hottest atomic/ionic gas in the bow shocks, are observable in the MIPS 24 μm band from the atomic/ionic line emissions. The dust emission from the protostar in the MIPS bands are good diagnostics of the circumstellar disks. However these low-surface-brightness features can be severely confused by the high brightness of the protostar and the disk. Deconvolution removes, or at least minimizes, the diffraction effects of the brightest features enabling improved visualization of the low-surface features around them. Furthermore, the resolution enhancement provides a sharper view of the outflow cavity walls, molecular jets and bow shocks.

4.2.1. Comparison with HST NIRCAM2 images

In this section we present the HiRes deconvolved Spitzer images of two outflow objects HH111 and HH300 at 3.6 μm and compare them with the high-resolution HST images at a closest wavelength of 2.2 μm observed by Reipurth et al. (1999) with the NICMOS2 camera (NIC2). The Spitzer and HST images of HH111 and HH300 are shown in Figures 8 and 9. The left panels in each figure show the Spitzer 3.6 μm (a) mosaic, and (b) deconvolved images. The right panels show the HST NIC2 F205W images (c) observed with full resolution (∼0.3''), and (d) smoothed to the resolution (∼0.65'') of the Spitzer deconvolved images. The protostar locations are marked by the arrows. The relative faintness of the protostars in the NIC2 image is due to heavy circumstellar obscuration. In both objects the reflection nebulosities (scattered light from the protostars) delineate the wide-angle outflow cavities around the protostars. The outflow morphologies are not at all evident in the Spitzer mosaic images, but are brought out more clearly in the HiRes deconvolved images due to both the increased resolution and the reduced sidelobe levels.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

HH111 shows a complex reflection nebula possibly due to the presence of two protostars. A second protostar "B" is observed in the HST and Spitzer images ∼3'' further to the west, which appears to contribute significantly to the illumination of the blueshifted outflow cavity of the HH 111 protostar "A" (Reipurth et al. 1997). The outflow lobes of HH111 (protostar "A") are roughly in the east–west direction and its eastern lobe is seen clearly without confusion from that of star "B." However, the simple outflow cone morphology of the western lobe is complicated by the north–south outflow lobes from the star "B." Considering the differences in their wavelengths and the brightness of the protostars between the HST NIC2 (Figure 8(d)) and Spitzer images (Figure 8(b)) we find that the HiRes deconvolution has retained a high fidelity preserving all the main features observed in the HST image.

In the HST image of HH300 (Figure 9(c)) a bright cometary reflection nebula surrounding the protostar is observed (Reipurth et al. 2000). The reflection nebula is remarkably symmetric about the jet axis toward the northeast from the protostar, representing the outflow lobe seen by scattered light from the star. Despite the diffraction pattern, it is clear that the reflection nebula shows part of the southwest lobe structure close to the star. The Spitzer mosaic image (Figure 9(a)) is dominated by the diffraction lobes of the bright protostar and there is only a slight hint of the presence of the northeast lobe. However, reducing the levels of the diffraction lobes with HiRes deconvolution it has been possible to bring out its presence more clearly. It may be noted that in this example we have used only 20 iterations, fewer than the nominal 50. Use of 50 iterations tends to break up the extended low-surface-brightness structure of the outflow lobe. The diffraction residue is present at a visible level but still low enough to bring out the underlying outflow morphology. It may be noted that the star is so bright that it was close to saturation in the data used for this image. (It was saturated in a number of BCDs of this region.) The choice of using 20 iterations at the expense of sacrificing resolution enhancement (see Figure 4) seems more appropriate for extended structures while with larger iterations deconvolution performs better for point sources and small-scale structures.

HH300 shows an example where HiRes deconvolution has been marginally successful in bringing out the extended structure around very bright point sources. In cases such as L1228 (Figure 13(a)) the deconvolution failed to bring out the low brightness wide-angle outflow morphology due to the presence of large levels of diffraction residue. Mismatches between the PRF and point-source image can result in large diffraction residue in the deconvolved image, especially when the sources are very bright, close to saturation, and they are not represented well by the PRF. On the other hand the HiRes deconvolution successfully brought out the outflow cone morphologies in L1448 IRS2 (Figure 11) where the protostar was not too bright.

4.2.2. Outflows in L1448-IRS2

Tobin et al. (2007) presented deep IRAC images that highlight the scattered light emission around many of the youngest protostars, the so-called Class 0 sources, in L1448. Here we present the results of HiRes processing of the IRAC and MIPS images. Figure 10 shows the IRAC 4.5 μm images of the full field containing L1448 mm, IRS2, and IRS3 and their large-scale jets and outflows. Note that the diffraction residuals around the bright stars in this field are negligible in the HiRes image. The marginal Airy ring residuals around the brightest star in the eastern edge of the image are due to the fact that this star is too bright and close to saturation. In Figure 11 we show a blow-up of the IRS2 protostar and outflow in all IRAC and MIPS 24 and 70 μm.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

4.2.3. Outflow in L1228

L1228 is a well-studied system with a wide-angle CO outflow and molecular H₂ jets. This object formed part of our HiRes analysis of First Look Survey (FLS) data and its results demonstrate the performance of the deconvolution on a star close to saturation. HiRes results are shown in Figures 12 and 13. In the mosaic images the emission around the protostar is overwhelmed by the diffraction lobes from the bright protostar. In the HiRes images the diffraction residuals are significantly low to bring out the features associated with the molecular jet. In Figure 13 we show a three-color image of IRAC emission and for comparison we reproduced the CO and H₂ molecular jets, from Arce & Sargent (2006). The outflow cavity in scattered light is not detected in the IRAC images presumably due to lower density in the envelope. The emissions at 4.5 μm as well as other IRAC bands show good correlation with the H₂ jet (seen as a string of knots).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The extragalactic component of the Spitzer FLS is composed of two parts—a large-area shallow survey covering an area of 2.5°×2° and a smaller ∼0.75°×0.3° strip, referred to as the verification survey (or FLSv) (see Frayer et al. 2004). Fryer et al. discuss the Spitzer counterparts in Very Large Array (VLA) and WSRT-FLSv sources (Condon et al. 2003; Morganti et al. 2004) representing radio selected Submillimeter Galaxies (SMGs). In Figure 14 we show the MIPS 24 μm images of their source no. 1. The HiRes deconvolution clearly presents a more clear view than the published 24 μm image (Figure 1 in Frayer et al. 2004) of the five individual component galaxies identified in the VLA/Westerbor/IRAC maps. Though the individual components are resolved in the mosaic image there is an indication of an extended envelope containing the five SMGs. However it is likely that the extended emission is the result of the sidelobes in the PRF. Indeed as demonstrated in Figure 14 HiRes deconvolution this envelope is removed gradually with each iteration as would be expected for sidelobes. At iteration 50 and above the diffraction envelope is removed entirely to the sensitivity limit and a sharper view of the components is obtained.

Zoom In Zoom Out Reset image size

The study of dusty debris disks around main-sequence stars can provide clues to the possible presence of orbiting bodies such as planets. In particular, by resolving the structures observed in the dust emission we can identify the grains trapped in various resonances caused by an unseen planet/planets around the star. One system of considerable interest is Fomalhaut (a Psα), which shows a ring morphology of the debris disk in submillimeter (Marsh et al. 2005) and in optical (Kalas et al. 2005) bands. The first Spitzer results at 24, 70, and 160 μm Early Release Observations (ERO) were presented by Stapelfeldt et al. (2004). They presented the first demonstration of HiRes deconvolution of MIPS data that was performed on the 70 μm images of Fomalhaut. Recently there have been new observations with Spitzer. In Figure 15 we show a new HiRes image at 70 μm that has been obtained using all the data available to date. The deconvolved image clearly brings out the pair of intensity maxima interpreted as the pericenter and apocenter of an inclined ring. The asymmetries in their intensities resulting primarily from the temperature differences and marginally due to dust column densities confirm the ringlike morphology but also show that the geometric center is displaced from the star by about 8 AU.

Zoom In Zoom Out Reset image size

In Figure 16 we show a HiRes deconvolution of the MIPS 24 μmimage of Tycho's supernova remnant. This demonstrates the fidelity of the HiRes deconvolution. A comparison of the deconvolved Spitzer image with Chandra X-ray images (Warren et al. 2005; Cassam-Chenaï et al. 2007) show that 24 μm IR emission shows an association with the blast wave, and not with the ejecta from the supernova as seen in the Kepler's supernova remnant (Blair et al. 2007). We are seeing ambient interstellar dust that has been swept up and heated by the main blast wave. The dusty grains are possibly conditionally heated by electrons in the X-ray-emitting hot gas. A combined analysis of deconvolved IR images of supernova remnants with X-ray images may provide a powerful diagnostic on plasma conditions in the remnant shells.

Zoom In Zoom Out Reset image size

The starbursts in interacting galaxies are traceable in MIPS bands and it is therefore necessary to compare MIPS images with those in the IRAC bands or optical bands. The Spitzer resolution in the MIPS bands are too coarse for a meaningful comparison. The HiRes deconvolved beams at 24 μm match the observed beams in the IRAC bands.

4.6.1. NGC2207 & IC2163

Elmegreen et al. (2006) presented the IRAC and MIPS observations of interacting galaxies NGC2207 & IC2163. In Figure 17 we show the IRAC 3.6 and 8 μm and MIPS 24 and 70 μm images. The HiRes deconvolution reveals detailed morphology of the interacting pair, particularly the starbursts as traced in the MIPS images. Of particular interest is the feature identified as "i" by Elmegreen et al. (2006) in the far-western edge. This is the brightest feature in the 24 and 70 μm images. This clump is also the brightest emission in H_α and radio continuum. The HiRes 24 μm image shows a ring-like resolved structure. Though this emission resembles the diffraction residue at the Airy rings, the emission is at a smaller radius and at a much higher level than that usually observed in point sources (Table 2).

Zoom In Zoom Out Reset image size

4.6.2. Arp 278

Arp 278 (NGC7253) is a starburst-interacting galaxy pair. The HiRes 8 and 24 μm images (Figure 18) bring out the prominence of the emission along the streamers.

Zoom In Zoom Out Reset image size

In Figures 19 and 20 we show the mosaic and deconvolved images of globular cluster NGC 4590 (M68) to demonstrate how well the HiRes deconvolution deblends the stars in such a high-density region. Nearly a factor of 3 enhancement in angular resolution (9 in area) reduces the area of the Airy disk of a star in the mosaic images by a factor of 9 in the deconvolved image allowing overlapping stars to deblend very well. The IRAC three-color images clearly show the dominance of stars in the shortest wavelength (blue). In HiRes images of IRAC channel 2 and 3 we can clearly identify more stars than in the corresponding mosaic images at least by factors of 3 (see Section 3.1). All of the seven giant stars observed by Lee et al. (2005) are clearly detected in all IRAC bands. One of the stars (No. 350), closest to the cluster center is blended in mosaic images while it is clearly resolved from the neighboring objects (as marked in Figure 20). It may be interesting to apply HiRes on the star clusters (total 92) discovered in the GLIMPSE survey (Mercer et al. 2005). Deconvolution could greatly improve upon the distribution of cluster memberships.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The Galactic center region is unique in our Galaxy and permits a close-up view of objects and phenomena that are impossible to obtain anywhere else. Because of the high stellar density toward the Galactic center, GLIMPSE is confusion limited. It is severely limited by blending of fainter sources and by the diffraction lobes of bright sources. In the HiRes images the sources are better resolved, and better isolated from the residual diffraction lobes. Use of HiRes deconvolution provides a means of reducing confusion and enabling detection of fainter sources, thereby increasing the number of detections by factors of 3 to 4 in some of the most scientifically interesting regions. As a test case we processed one of the GLIMPSE II (see Meade et al. 2007) AORs imaging a 1.7°× 0.37° field in the inner Galaxy. In Figure 21 we show the mosaic (made using the SSC BCDs) and HiRes images of a 10 (arcmin)² region centered near G359.67+0.633 at 4.5 and 8 μm bands. These images demonstrate how HiRes resolves point sources well below the confusion limit in the mosaic images. Clearly we can see many more sources in the HiRes image that resemble noise in the mosaic. In fact SExtractor finds four times more sources above 5σ in the HiRes images than in the mosaic. Furthermore the presence of a bright source (saturated at 4.5 μm) and its associated effects such as "muxstriping" (8 μm) as seen in mosaic images has not affected HiRes performance in the adjacent areas of the images. In this particular example the diffraction lobes around the bright star are not cleaned well in the HiRes image because of the saturation.

Zoom In Zoom Out Reset image size

In Figure 22 we show an example of the source count as extracted using SExtractor in a 5' × 5' area around the region shown in Figure 21. The apparent excess of sources in the mosaic image at the highest flux is due to the blending of multiple sources identified as single objects with a higher flux density while in the deconvolved image they are identified correctly with appropriate lower fluxes.

Zoom In Zoom Out Reset image size