vil_nitf2_image.h

Go to the documentation of this file.

//:
// \file
// vil_nitf2: Written by Rob Radtke (rob@) and Harry Voorhees (hlv@) of
// Stellar Science Ltd. Co. (stellarscience.com) for
// Air Force Research Laboratory, 2005.

#ifndef VIL_NITF2_IMAGE_H
#define VIL_NITF2_IMAGE_H

#include <vector>
#include <vil/vil_blocked_image_resource.h>

#ifdef _MSC_VER
#  include <vcl_msvc_warnings.h>
#endif
#include <cassert>

#include <vil/vil_stream.h>

#include "vil_nitf2_image_subheader.h"
#include "vil_nitf2_header.h"

#include <vil/vil_file_format.h>

class vil_nitf2_des;

class vil_nitf2_file_format : public vil_file_format
{
 public:
  char const *tag() const override;
  vil_image_resource_sptr make_input_image(vil_stream *vs) override;
  vil_image_resource_sptr make_output_image(vil_stream* vs,
                                                    unsigned nx,
                                                    unsigned ny,
                                                    unsigned nplanes,
                                                    enum vil_pixel_format) override;
};

//: Class for reading NITF 2.1 imagery files.
// It works just like any other vil_image_resource class except that it does
// support retrieving multiple images from the same file.
// Call nimages() to find out how many images are in this resource, and then call
// set_current_image() to tell the class which image you want to work with presently.
// All regular vil_image_resource functions will operate on the current image.
//
// You can access image header information through get_image_headers() and file
// header information through get_header().
//
// Known reading limitations:
//  - The only supported compression schemes are uncompressed (all 4 different
//    types of data layouts: band sequential, band interleaved, band interleaved
//    by row, and band interleaved by pixel) and JPEG 2000 compression (via plugin).
//  Other forms of compression do not work (eg. JPEG (regular, non-j2k)
//  - Writing is unsupported at this time
//  - graphic segments are not read in
//  - text segments are not read in
//  - The class was designed to handle images with PVTYPE=C (complex), but images of
//    this type are completely untested.
//
// Things that do work:
//  - Reading uncompressed and image data (ints, shorts, 8 bit int, floats etc)
//  - parsing and validating file headers, and image headers
//  - enum_data masks (and pad pixels) are supported through vil_nitf2_data_mask_table
//  - PJUST = L
//  - Images with LUTS will read in correctly, but you will need to apply the LUT
//    yourself.  You can access the lut information for p-th band of the current image by calling
//    current_image_header()->get_lut_info(p, out_n_luts, out_ne_lut, out_lut_data)
//  - The class does not have built-in support for JPEG2000 compressed NITF files
//    Using cmake though, you can configure VXL to link against Er Mapper's freely
//    available ECW JPEG2000 SDK (http://ermapper.com/downloads/sdks.aspx#16).  If you
//    do that, then this class will automatically be able to read NITF files that are
//    JPEG 2000 compressed.
//
class vil_nitf2_image : public vil_blocked_image_resource
{
 public:
  //: Instantiate an image resource, but doesn't read anything.
  // You'll want to call parse_headers() before you do anything with me
  // (eg. before you ask for any image data).
  // If that returns false, then I am invalid and useless to you in every way.
  vil_nitf2_image( vil_stream* is );
  vil_nitf2_image( const std::string& filePath, const char* mode );

  ~vil_nitf2_image() override;

  //:return the image info of the current image
  unsigned nplanes() const override;
  unsigned ni() const override;
  unsigned nj() const override;
  enum vil_pixel_format pixel_format () const override;

  //: Block size in columns
  unsigned size_block_i() const override;

  //: Block size in rows
  unsigned size_block_j() const override;

  //: Number of blocks in image width
  unsigned n_block_i() const override;

  //: Number of blocks in image height
  unsigned n_block_j() const override;

  //: returns "nitf vM.N"
  char const * file_format() const override;


  // is the current image JPEG 2000 compressed
  bool is_jpeg_2000_compressed() const;

  // This function is sort of a hack.  It will return a decimated version
  // of the region specified by (i0,j0){i0+ni,j0+nj).  It will be decimated by i_factorXj_factor.
  // That is the image returned will have size (ni/i_factor, nj/j_factor).
  // The reason this function is a hack is because it only works for jpeg 2000 compressed
  // NITF files (that is, is_jpeg_2000_compressed() returns true).  You would normally want to use
  // vil_decimate_image_resource to obtain decimated images.  That will work for jpeg2000
  // compressed NITF files too, but that class/function doesn't take advantage of the JPEG 2000
  // spec's built in optimization for obtaining decimated versions of large images.  This function
  // does.  Well, more accurately, it can.  It depends on the implementation of the s_decode_jpeg_2000()
  // function you've provided.
  virtual vil_image_view_base_sptr get_copy_view_decimated_j2k( unsigned i0, unsigned ni,
                                                                unsigned j0, unsigned nj,
                                                                double i_factor, double j_factor ) const;
  virtual vil_image_view_base_sptr get_copy_view_decimated_j2k( double i_factor, double j_factor ) const
  { return get_copy_view_decimated_j2k( 0, ni(), 0, nj(), i_factor, j_factor ); }

  vil_image_view_base_sptr get_copy_view(unsigned i0, unsigned ni,
                                                 unsigned j0, unsigned nj) const override;
  virtual vil_image_view_base_sptr get_copy_view( ) const
  { return get_copy_view( 0, ni(), 0, nj() ); }

  bool put_view (const vil_image_view_base& /* im */, unsigned /* i0 */, unsigned /* j0 */ ) override
  { return false; }

  bool put_block(unsigned  /*block_index_i*/, unsigned  /*block_index_j*/,
                         const vil_image_view_base& /*blk*/ ) override
  { return false; }

  vil_image_view_base_sptr get_block( unsigned int blockIndexX, unsigned int blockIndexY ) const override;

  bool get_property (char const *tag, void *property_value=nullptr) const override;

  //const vil_nitf2_header& getFileHeader() const;
  const std::vector< vil_nitf2_image_subheader* >& get_image_headers() const
  { return m_image_headers; }
  const vil_nitf2_header& get_header() const
  { return m_file_header; }
  const std::vector< vil_nitf2_des* >& get_des() const
  { return m_des; }

  //:
  //  Since the VIL API (eg. get_view()) for retrieving image data
  //  doesn't support files with multiple images, clients will
  //  need to call this function to tell get_view() which image to read in.
  //
  //  Overloaded from vil_image_resource.  The NITF 2.x file format does support
  //  multiple images per file and you can use this API to access each and every
  //  one of them.
  //
  //  Note: By default, the first image is always used.  If you don't call this
  //  function at all, then you will only see the first image in a given file.
  virtual void set_current_image( unsigned int index );
  virtual unsigned int current_image() const;
  virtual unsigned int nimages() const;

  bool parse_headers();
  vil_nitf2_classification::file_version file_version() const;

  //:
  // All instances of vil_nitf2_image will use s_decode_jpeg_2000() to decode
  // JPEG 2000 streams if you set the function.  If unset, then the library
  // will not be able to read JPEG 2000 compressed NITF files.
  //
  static vil_image_view_base_sptr ( *s_decode_jpeg_2000 )( vil_stream* vs,
                                                           unsigned i0, unsigned ni,
                                                           unsigned j0, unsigned nj,
                                                           double i_factor, double j_factor );

  // I allocate the return value, but you own it after I return it to you
  // so you need to delete it.
  virtual vil_nitf2_field::field_tree* get_tree() const;

 protected:
  virtual vil_image_view_base_sptr get_block_j2k( unsigned int blockIndexX, unsigned int blockIndexY ) const;
  virtual vil_image_view_base_sptr get_copy_view_uncompressed(unsigned i0, unsigned ni,
                                                              unsigned j0, unsigned nj) const;


  // Returns the offset (in bytes) from the beginning of the NITF file
  // to the beginning of the specified portion of the NITF stream.  For example:
  //
  // get_offset_to( enum_text_segments, Header, 0) will return the offset from the
  // beginning of the NITF stream to the beginning of the first text segment's
  // subheader.  You'd better make sure there is at least one text segment before
  // you call this.
  //
  // get_offset_to( enum_image_segments, enum_data, 3) will return the offset from the
  // beginning of the NITF stream to the beginning of the fourth image segment's
  // data section.  You'd better make sure there is at least four image segments before
  // you call this.
  vil_streampos get_offset_to( vil_nitf2_header::section_type sec, vil_nitf2_header::portion_type por, unsigned int index = 0 ) const;
  // Returns the offset (in bytes) from the beginning of section_type 'sec' and the beginning of the specified portion
  // (subheader or data).  If more then one of these segments exist, then use 'index' to select which one you want.
  vil_streampos size_to( vil_nitf2_header::section_type sec, vil_nitf2_header::portion_type por, int index ) const;

  // Returns the overall offset to the specified image/block/band combination.
  // If this block isn't present in the stream (ie. it's all blank), then
  // I'll return 0;
  vil_streampos get_offset_to_image_data_block_band( unsigned int imageIndex,
                                                     unsigned int blockIndexX,
                                                     unsigned int blockIndexY,
                                                     int bandIndex ) const;

  //main file header
  vil_nitf2_header m_file_header;
  //image header(s)
  std::vector< vil_nitf2_image_subheader* > m_image_headers;
  void clear_image_headers();
  const vil_nitf2_image_subheader* current_image_header() const;
  //DESs (if any)
  std::vector< vil_nitf2_des* > m_des;
  void clear_des();

  vil_stream* m_stream;
  unsigned int m_current_image_index;
};

//: This function does a lot of work for \sa byte_align_data().
// It will strip one value of ni bits and return it as a value of type T
// (with zero padding on the MSBs).
// Both io and ni are lengths (in bits - not bytes).
//
// \param i0: Offset (in bits - not bytes) from in_val[0].  This will be the start
//            of the bits extracted from in_val.
// \param ni: number of bits (starting from i0) that will be extracted from in_val.
template< class T >
T get_bits( const T* in_val, unsigned int i0, unsigned int ni )
{
  unsigned int sample_offset = i0 / ( sizeof(T)*8 );
  unsigned int bit_offset = i0 % ( sizeof(T)*8 );

  unsigned int strip_left = bit_offset;
  int strip_right = ( sizeof( T ) * 8 ) - ( bit_offset + ni );
  T temp = in_val[sample_offset];
  if ( strip_left > 0 ){
    //strip off the appropriate bits from the std::left (replacing them with zeros)
    temp = temp << strip_left;
    temp = temp >> strip_left;
  }
  if ( strip_right > 0 ){
    //strip off the appropriate bits from the std::right
    //the bit shift operator wasn't having the correct effect, so that'w
    //why the for loop
    for ( int i = 0 ; i < strip_right ; i++ ) temp /= 2;
    //temp = temp >> strip_right;
  }
  else if ( strip_right < 0 ){
    //we didn't have enough bits in the first element of the in_val array
    //need to get some from the next element
    for ( int i = 0 ; i < (-strip_right) ; i++ ) temp *= 2;
    temp += get_bits<T>( in_val+sample_offset+1, 0, -strip_right );
  }
  return temp;
}

//: This function will byte align the data in in_data and store the result in out_data.
// For example, let's say that you had in_data is of type unsigned char
// and contains the following data: 110010111001111010000110.
// In other words:
// in_data[0] = 203 (11001011)
// in_data[1] = 158 (10011110)
// in_data[2] = 134 (10000110)
// Let's further say you called this function like this: byte_align_data( in_data, 8, 3, out_data ).
// Then, when the function finished, out_data would look like this:
// out_data[0] = 6 (00000110)
// out_data[1] = 2 (00000010)
// out_data[2] = 7 (00000111)
// out_data[3] = 1 (00000001)
// out_data[4] = 7 (00000111)
// out_data[5] = 2 (00000010)
// out_data[6] = 0 (00000000)
// out_data[7] = 6 (00000110)
//
// Basically, what the function did was group the bitstream into groups of three and then store all of the
// values into out_data.  It had to zero pad all the values (on the MSB side) to get them into out_data.  That's
// why out_data is bigger.
//
// This function works with other unsigned types of data too.  For example, let's say in_data was of type unsigned int
// and contained the following bits: 0100110010111000 0111101100000000 1111000011110000 (note that this bitstream is shown
// in big endian notation, that will not be the case if you are on a little endian machine -- this is just for illustration)
// in other words:
// in_data[0] = 19640 (0100110010111000) [shown in big endian for illustrative purposes only]
// in_data[1] = 31488 (0111101100000000) [shown in big endian for illustrative purposes only]
// in_data[2] = 61680 (1111000011110000) [shown in big endian for illustrative purposes only]
// Let's further say, you called this function like this byte_align_data( in_data, 4, 12, out_data ).
// Then out_data would be aligned along two byte (sixteen bit) boundaries and would look like this:
// out_data[0] = 1227 (0000010011001011) [shown in big endian for illustrative purposes only]
// out_data[1] = 2171 (0000100001111011) [shown in big endian for illustrative purposes only]
// out_data[2] = 15   (0000000000001111) [shown in big endian for illustrative purposes only]
// out_data[3] = 240  (0000000011110000) [shown in big endian for illustrative purposes only]
//
// Because of the fact that this function uses bit shifting operators, and the behavior of the std::right
// shift operator is implementation specific when applied to a negative number, you should probably
// only use this function on unsigned data.
//
// \param in_data: The input data.  It must be at least (num_samples*in_bits_per_sample\8) bytes long.
//                The values should have the endianness of your platform.
// \param num_samples: The number of actual samples in in_data.
// \param in_bits_per_sample: The bits per sample in in_data
// \param out_data: I'll store the output data here.  It must be at least (num_samples*sizeof(T)) bytes long
//                 The values will have the endianness of your platform.
//
// Note that inBitsPerSampe must not be >= sizeof(T).  If they were to be equal, then this function
// would have nothing to do (in_data is already byte aligned).  If in_bits_per_sample were less than sizeof(T),
// then each element of out_data would be too small to store the corresponding elements in in_data.
//
// Note that there is a specialization for the bool case which just casts it to an 8 bit quantity then calls
// this same function.  This is because the logic in get_bits<> doesn't work for the bool case.
template< class T >
T* byte_align_data( T* in_data, unsigned int num_samples, unsigned int in_bits_per_sample, T* out_data )
{
  assert( in_bits_per_sample < sizeof(T)*8 );

  //grab each value from the bitstream (in_data) that we need... one at a time
  unsigned int bit_offset = 0;
  for ( unsigned int o = 0 ; o < num_samples ; o++ ){
    out_data[o] = get_bits<T>( in_data, bit_offset, in_bits_per_sample );
    bit_offset+=in_bits_per_sample;
  }

  return out_data;
}

template<> bool* byte_align_data<bool>( bool* in_data, unsigned int num_samples, unsigned int in_bits_per_sample, bool* out_data );

#endif // VIL_NITF2_IMAGE_H