[c @sddlZddlZddlZddlZddlZddlZddlZddlm Z m Z m Z m Z ddl mZddlmZddlmZddlmZmZmZdd lmZdd lmZdd lmZdd lmZmZm Z m!Z!d dl"m#Z#m$Z$d dl%m&Z&d dl'm(Z(d dl)m*Z*d dl+m,Z,m-Z-yddl.m/Z/e0Z1Z2Wne3k re4Z1Z2nXdZ5dZ6dZ7ide56de66de76Z8dZ9dZ:dZ;dZ<e5Z=e9Z>dZ?dZ@dZAd ZBiZCe1ry>e/jDdkZEe/jDdkZFe/jDdkr^deCd^[Zz][a-zA-Z]*)(?P[1-9][0-9 ]*$)?RRR"tZQUANTIZtZDITHER0tZTILEtZNAMEtZVALcCsU|dkrt}n|j|_|j|_|j|_|j|_||_dS(N(tNoneR t_cardst_keyword_indicest _rvkc_indicest _modifiedt _table_header(tselft table_headert image_header((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyt__init__ls      cCs|j|||rdSt|tr@|j|\}}n+t|tr^|\}}n |d}}|j|r~dStt|j |||dk r|j |}||j ||f`_. Basically, the compressed image tiles are stored in rows of a variable length array column in a FITS binary table. The astropy.io.fits recognizes that this binary table extension contains an image and treats it as if it were an image extension. Under this tile-compression format, FITS header keywords remain uncompressed. At this time, Astropy does not support the ability to extract and uncompress sections of the image without having to uncompress the entire image. The astropy.io.fits package supports 3 general-purpose compression algorithms plus one other special-purpose compression technique that is designed for data masks with positive integer pixel values. The 3 general purpose algorithms are GZIP, Rice, and HCOMPRESS, and the special-purpose technique is the IRAF pixel list compression technique (PLIO). The ``compression_type`` parameter defines the compression algorithm to be used. The FITS image can be subdivided into any desired rectangular grid of compression tiles. With the GZIP, Rice, and PLIO algorithms, the default is to take each row of the image as a tile. The HCOMPRESS algorithm is inherently 2-dimensional in nature, so the default in this case is to take 16 rows of the image per tile. In most cases, it makes little difference what tiling pattern is used, so the default tiles are usually adequate. In the case of very small images, it could be more efficient to compress the whole image as a single tile. Note that the image dimensions are not required to be an integer multiple of the tile dimensions; if not, then the tiles at the edges of the image will be smaller than the other tiles. The ``tile_size`` parameter may be provided as a list of tile sizes, one for each dimension in the image. For example a ``tile_size`` value of ``[100,100]`` would divide a 300 X 300 image into 9 100 X 100 tiles. The 4 supported image compression algorithms are all 'lossless' when applied to integer FITS images; the pixel values are preserved exactly with no loss of information during the compression and uncompression process. In addition, the HCOMPRESS algorithm supports a 'lossy' compression mode that will produce larger amount of image compression. This is achieved by specifying a non-zero value for the ``hcomp_scale`` parameter. Since the amount of compression that is achieved depends directly on the RMS noise in the image, it is usually more convenient to specify the ``hcomp_scale`` factor relative to the RMS noise. Setting ``hcomp_scale = 2.5`` means use a scale factor that is 2.5 times the calculated RMS noise in the image tile. In some cases it may be desirable to specify the exact scaling to be used, instead of specifying it relative to the calculated noise value. This may be done by specifying the negative of the desired scale value (typically in the range -2 to -100). Very high compression factors (of 100 or more) can be achieved by using large ``hcomp_scale`` values, however, this can produce undesirable 'blocky' artifacts in the compressed image. A variation of the HCOMPRESS algorithm (called HSCOMPRESS) can be used in this case to apply a small amount of smoothing of the image when it is uncompressed to help cover up these artifacts. This smoothing is purely cosmetic and does not cause any significant change to the image pixel values. Setting the ``hcomp_smooth`` parameter to 1 will engage the smoothing algorithm. Floating point FITS images (which have ``BITPIX`` = -32 or -64) usually contain too much 'noise' in the least significant bits of the mantissa of the pixel values to be effectively compressed with any lossless algorithm. Consequently, floating point images are first quantized into scaled integer pixel values (and thus throwing away much of the noise) before being compressed with the specified algorithm (either GZIP, RICE, or HCOMPRESS). This technique produces much higher compression factors than simply using the GZIP utility to externally compress the whole FITS file, but it also means that the original floating point value pixel values are not exactly preserved. When done properly, this integer scaling technique will only discard the insignificant noise while still preserving all the real information in the image. The amount of precision that is retained in the pixel values is controlled by the ``quantize_level`` parameter. Larger values will result in compressed images whose pixels more closely match the floating point pixel values, but at the same time the amount of compression that is achieved will be reduced. Users should experiment with different values for this parameter to determine the optimal value that preserves all the useful information in the image, without needlessly preserving all the 'noise' which will hurt the compression efficiency. The default value for the ``quantize_level`` scale factor is 16, which means that scaled integer pixel values will be quantized such that the difference between adjacent integer values will be 1/16th of the noise level in the image background. An optimized algorithm is used to accurately estimate the noise in the image. As an example, if the RMS noise in the background pixels of an image = 32.0, then the spacing between adjacent scaled integer pixel values will equal 2.0 by default. Note that the RMS noise is independently calculated for each tile of the image, so the resulting integer scaling factor may fluctuate slightly for each tile. In some cases, it may be desirable to specify the exact quantization level to be used, instead of specifying it relative to the calculated noise value. This may be done by specifying the negative of desired quantization level for the value of ``quantize_level``. In the previous example, one could specify ``quantize_level = -2.0`` so that the quantized integer levels differ by 2.0. Larger negative values for ``quantize_level`` means that the levels are more coarsely-spaced, and will produce higher compression factors. The quantization algorithm can also apply one of two random dithering methods in order to reduce bias in the measured intensity of background regions. The default method, specified with the constant ``SUBTRACTIVE_DITHER_1`` adds dithering to the zero-point of the quantization array itself rather than adding noise to the actual image. The random noise is added on a pixel-by-pixel basis, so in order restore each pixel from its integer value to its floating point value it is necessary to replay the same sequence of random numbers for each pixel (see below). The other method, ``SUBTRACTIVE_DITHER_2``, is exactly like the first except that before dithering any pixel with a floating point value of ``0.0`` is replaced with the special integer value ``-2147483647``. When the image is uncompressed, pixels with this value are restored back to ``0.0`` exactly. Finally, a value of ``NO_DITHER`` disables dithering entirely. As mentioned above, when using the subtractive dithering algorithm it is necessary to be able to generate a (pseudo-)random sequence of noise for each pixel, and replay that same sequence upon decompressing. To facilitate this, a random seed between 1 and 10000 (inclusive) is used to seed a random number generator, and that seed is stored in the ``ZDITHER0`` keyword in the header of the compressed HDU. In order to use that seed to generate the same sequence of random numbers the same random number generator must be used at compression and decompression time; for that reason the tiled image convention provides an implementation of a very simple pseudo-random number generator. The seed itself can be provided in one of three ways, controllable by the ``dither_seed`` argument: It may be specified manually, or it may be generated arbitrarily based on the system's clock (``DITHER_SEED_CLOCK``) or based on a checksum of the pixels in the image's first tile (``DITHER_SEED_CHECKSUM``). The clock-based method is the default, and is sufficient to ensure that the value is reasonably "arbitrary" and that the same seed is unlikely to be generated sequentially. The checksum method, on the other hand, ensures that the same seed is used every time for a specific image. This is particularly useful for software testing as it ensures that the same image will always use the same seed. shThe astropy.io.fits.compression module is not available. Creation of compressed image HDUs is disabled.s@Keyword argument {} to {} is pending deprecation; use {} insteadtquantize_methodt dither_seedtdatatheaderR!iitBZEROtBSCALER N(!tCOMPRESSION_SUPPORTEDt ExceptiontCMTYPE_ALIASESR}tDEPRECATED_KWARGStitemsRpRqRWt __class__RRtlocalsRRKRRDR;Rt_update_header_datat_do_not_scale_image_datat_uintt _scale_backRt_headertstrt_axest_bzerot_bscalet_bitpixt _orig_bzerot _orig_bscalet _orig_bitpix(RARRtnameRRRRRRRtdo_not_scale_image_datatuintt scale_backtkwargstcompression_optstoldargtnewargtaxis((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyRDsB      "    K    cCs|jd}|jdkr tS|j}t|trG|j}n|dkrWtSd|ksn|d rrtStrtrt Stst j dt tStSdS(NiR(tBINTABLEtA3DTABLERsFailure matching header to a compressed image HDU: The compression module is not available. The HDU will be treated as a Binary Table HDU.(RR( tcardsRPR_RORFRtrstripRtCOMPRESSION_ENABLEDRbRpRqR(RyRRXtxtension((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyt match_headers"     c 5 Cstd|jd|j} t|j| j|_| j|_~ |jr|jjdk} | rt rt dj |j |j f|jjqnt} | rd|jkrd}n|r|jjd|dd d ||_ n|jd|_ |rP|dkr1tjdj ttt}n|jjd|dd d n'|jjdt}tj||}|r|jdd} |jdd} d}| dkr|jjd| d |d}n| dkr|jjd| d |n|jd}|jd}n d}d}|jjdddd d |dkrc| rZdnd}n| rod nd!}|jjd"|d#d dtd$|jdd%|}|jd}|d&kr|dkrd'}trd(}| rd nd!}nd)}|d*kr | rd nd!}nu|d+krA| r8dnd}nT|dkrb| rYd,nd-}n3|d.kr| rzd/nd0}n| rd1nd2}|jjd3|d4d d"|jjd5|d#d d3td$|d%|}|jjd6d7d8d d5|jjd9d:d;d d6td$|jd6d%|jd9}|jjd<d=d>d d9|jjd?d:d;d d<d?}td$|jd<d%|jd?}t||||g}nfd@}d"}d3d5d6d9d<d?g}x0|D](}y|j|=Wqtk r qXqWt|g}|jjdA|jjdB|jjd |dCd dD|jjdEtdFd ||jjdG||d dE|jjdH|jd|d dGd@}xUtry6|jdHt|=|jdIt|=|d@7}Wqtk r PqXqW|jd}|s,g}n+t||krWtjdJtg}n|dkre |jdAd'ks|jdKd'krt dLn~|r|d&d'ks|d@d'krt dMntg|D]} | d@kr| ^q}!|!dkrt dNqn|r|d&d&kr|d@d&kr|jdA|d&<|jdK|d@= 3.35tEXTNAMEtCOMPRESSED_IMAGEs#name of this binary table extensionR`RmRtGZIP_1tGZIP_2tPLIO_1t HCOMPRESS_1sBUnknown compression type provided. Default ({}) compression used.Rscompression algorithmRgRg?R)R*sdata type of original imagesdimension of original imagetTTYPE1tCOMPRESSED_DATAslabel for field 1t1QIt1PIt1QBt1PBtTFORM1s+data format of field: variable length arrayRRWiitGZIP_COMPRESSED_DATAtUNCOMPRESSED_DATAiit1QJt1PJit1QEt1PEt1QDt1PDtTTYPE2slabel for field 2tTFORM2tTTYPE3tZSCALEslabel for field 3tTFORM3t1Ds#data format of field: 8-byte DOUBLEtTTYPE4tZZEROslabel for field 4tTFORM4itNAXIS1swidth of table in bytessnumber of fields in each rowR1Rs#extension contains compressed imageR R!R8sQProvided tile size not appropriate for the data. Default tile size will be used.tNAXIS2s-Hcompress minimum image dimension is 4 pixelss,Hcompress minimum tile dimension is 4 pixelssmHCOMPRESS can only support 2-dimensional tile sizes.All but two of the tile_size dimensions must be set to 1.iiiiiiiiiis4Last tile along 1st dimension has less than 4 pixelss4Last tile along 2nd dimension has less than 4 pixelstZNAXIS1slength of original image axisssize of tiles to be compressedsnumber of rows in tableR9R:tNOISEBITsSCALE sSMOOTH tZNAME1t BLOCKSIZEscompression block sizetZVAL1spixels per blocktZNAME2tBYTEPIXsbytes per pixel (1, 2, 4, or 8)tZVAL2tSCALEsHCOMPRESS scale factortSMOOTHsHCOMPRESS smooth options!floating point quantization levels@Unknown quantization method provided. Default method ({}) used.s No dithering during quantizationsPixel Quantization AlgorithmR6R7s'dithering offset when quantizing floatsR'R#RhR+R%R-R,R(R$tIMAGER/R.R0R3R2R5R4NI(RRRRR(6RRRR&Rt _image_headerRt _has_datatnbytestCFITSIO_SUPPORTS_Q_FORMATtIOErrorRWRtverR_RRpRqtDEFAULT_COMPRESSION_TYPERR}RtcommentsRtCFITSIO_SUPPORTS_GZIPDATARRVtdtypetitemsizeRbRRaR\RR]t enumeratetcolumnsR;tDEFAULT_QUANTIZE_LEVELtDEFAULT_HCOMP_SCALEtDEFAULT_BLOCK_SIZEtDEFAULT_BYTE_PIXRRRtQUANTIZE_METHOD_NAMEStDEFAULT_QUANTIZE_METHODRFRRRtt_generate_dither_seedtDEFAULT_DITHER_SEEDt _countblanks(5RARCRRRRRRRRt image_hduthuge_hdutbzerotbscalet after_keywordtbitpix_commentt naxis_commentttform1tcol1tzbitpixtncolstttype2ttform2tcol2tcol3R`tcol4tcolst to_removetkRtnaxisttst major_dimstitnaxis2tdimtremaint last_znaxistafter1tnrowsRtznaxistztiletznametzvaltbytepixtzquantiz_commenttvtrequired_blankst image_blankst table_blankst_((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyRs@   '                                 + & (                                                                 cCstj|}|dkr|S|jdks=|jdkr|j}tj|d|}d}d|jj j kr|jd}n\d|j krtj|j ddd}n.d|j krtj|j ddd}n|dk r||k}n|j dkr(tj ||j |n|jdkrYtj||jd|dd n|dk rtj|tj|}qn|j|j|S( NiiRtZBLANKtint32tBLANKtouttcastingtunsafe(Rtdecompress_hduR;RRt_dtype_for_bitpixtnptarraytcompressed_dataRtnamesRRtmultiplyRtaddtwheretnant_update_header_scale_infoR(RARt new_dtypetzblanktblanks((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyR_s.   " cCs\|dk rXt|tj s1|jjdk rXtdjt||jjndS(Ns:CompImageHDU data has incorrect type:{}; dtype.fields = {}( R;RFR3tndarrayRtfieldst TypeErrorRWttype(RAR((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyRs cCsFtt|j}t|tjjr5|jd=|S|j|j S(NR( RKRRRFR3trectrecarrayt__dict__t_update_compressed_dataR5(RAR5((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyR5s   cCs7d|jkr3|jd`|jd=tjndS(NR5(REt_coldefstgctcollect(RA((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyR5s  cCstt|jS(sX Shape of the image array--should be equivalent to ``self.data.shape``. (RItreversedR(RA((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pytshapesc Cs5t|dr|jS|jj}x3t|D]%}tj|dtr2||=q2q2Wd|jkr|jd|jd|jjdddnqd|jkr|jddkrt j d t n|jd d|jjdddn|jd ddd|jd |jd |jjd dd |jd|jd|jjdddd}xft |dD]T}dt |d }|d }|j||j||jj|d||}qsW|d}xYt|dD]G}yt|d}Wntk rqnX||kr||=qqWd|jkrm|jd|jd|jjdd|n|jddd|d|jkr|jd|jd|jjdddn|jdd ddd|jkr |jd|jd|jjdnd|jkrA|jd|jd|jjdnd|jkrw|jd|jd|jjdnd|jkr|jd|jd|jjdnd |jkr|jd d!kr|d =n|jj}|j} x"t || D]} |jqWt|j||_|jS("NRRqR#R'RhiR$Rs3ZTENSION keyword in compressed extension != 'IMAGE'R(R)R iR*R!iR`sNAXIS?*iR.R/R0R1R%R+R,R-R2R3R4R5RR(thasattrRRtcopyRR&RJR_RRpRqRRRtlistRGRRR]( RARCRPt last_naxisRR RtnR)R(R*((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyRs             cCs|jj}|jr|jdkr3d\}}qt|jj}|jjj}|j t |}||j dd}nWd }x=t |j dD](}||j dt|df7}qWt|j d}|j|j|t|j ||fS( sC Summarize the HDU: name, dimensions, and formats. tt.iR*R)N(((RQ((RRt _data_loadedRR;RNRKRRtreverseRItrfindRRRRRRa(RAt class_namet_shapet_formatR((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyt_summary:s    &c Cst|jjj}||jks7|jj|jkrJ|j|jn|j}t|jjrt j |jt |jjddj |jjj |_t}n|jjj }|r|jjjr|jjtq|jjt|_nzr|jd}|jd|}d|jdRB("RRRRRRKRRR R3R4RRWRR_tisnativetflagst writeabletbyteswapRbRt_theapR5Rt compress_hduRt newbyteordertviewRCRDR RGt _heapoffsett _heapsize( RAt image_bitpixtold_datat should_swapRttbsizetheapsizeRtbufR5((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyRFZsD$        toldiic Cs|jdkrdS|dkr/t|j}ntt|}|dksV|dkre|}|}n|dkr|j}|j}n|dkr7t|tj rd}d}q7tj j |jj }tj j |jj } |tjkr |}| |d}q7| |d}| |dd|jd}n|dkrutj|j|d |jd d ||jd kr^|Sq^Wn|dkrtjdS|d krtjd Sd S( s Determine the dtype that the data should be converted to depending on the BITPIX value in the header, and possibly on the BSCALE value as well. Returns None if there should not be any change. iituint16i tuint32i@tuint64tfloat64itfloat32N(RRRR3RR(RAtbitpixtbitsR((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyR2Ss "#    cCs|j r|jdko%|jdk rxUddgD]G}x>|j|jfD]*}tt||=|jWdQXqOWq6W|dkr|j }n|dk rt |j |jdsRRtcss|]}t|VqdS(N(RG(Rtx((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pys s(RRAR\RWtDITHER_SEED_CHECKSUMRRRTRRIRctsumtctypestc_ulongROtDITHER_SEED_CLOCKtmathtmodfttimetid(RAtseedRRt tile_dimst first_tiletcsum((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyRs$   0  , ($RRRRRbt_manages_own_heapR;RRtDEFAULT_HCOMP_SMOOTHRRRR_RDRRRRRtsetterR5tdeletertpropertyRKRRYRFR~RRRRR2R;R(((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pyRzsX   + x El$   (LRRHRRRRptnumpyR3tbaseRRRRtimageRttableRRXRtcolumnRRR R RutfitsrecR RR tutilR RRRt extern.sixRRtextern.six.movesRtutilsRt utils.compatRtutils.exceptionsRRRQRRbRRt ImportErrorR_RRRRRRRRRRRRRRRtCFITSIO_VERSIONRRtAttributeErrorRtCOMPRESSION_KEYWORDSR&R(((s=lib/python2.7/site-packages/astropy/io/fits/hdu/compressed.pytsr       ""    +