o jH@sZdZddlZddlZddlZddlmZddlmZddlm Z m Z m Z GdddZ dS) aExtract images from PDF. Both raster images and vector graphics are considered: * Normal images like jpeg or png could be extracted with method ``page.get_text('rawdict')`` and ``Page.get_images()``. Note the process for png images with alpha channel. * Vector graphics are actually composed of a group of paths, represented by operators like ``re``, ``m``, ``l`` and ``c``. They're detected by finding the contours with ``opencv``. N) Collection) BlockType)recursive_xy_cutinner_contoursxy_project_profilec@seZdZdZdejddfddZ d+d ejd ed e fd d Z   d+d ejd ede fddZ e de fddZd,de fddZde de de de fddZe dejd ejfddZe dejde fd d!Ze dejd"ed efd#d$Ze d%ejd&efd'd(Ze dejfd)d*ZdS)-ImagesExtractorzExtract images from PDF.pagereturnNcCs ||_dS)zoExtract images from PDF page. Args: page (fitz.Page): pdf page to extract images. N)_page)selfr r _/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/pdf2docx/image/ImagesExtractor.py__init__s zImagesExtractor.__init__F@bboxrm_imagezoomc Cs|j|jd|d}|dur|jj}n |jjr||jj}n|}|jj|@}t||}|jj||d}|jj}| D] \} } | | | q;|S)aClip page pixmap according to ``bbox``. Args: bbox (fitz.Rect, optional): Target area to clip. Defaults to None, i.e. entire page. Note that ``bbox`` depends on un-rotated page CS, while clipping page is based on the final page. rm_image (bool): remove images or not. zoom (float, optional): Improve resolution by this rate. Defaults to 3.0. Returns: fitz.Pixmap: The extracted pixmap. T)rm_textrN)Zclipmatrix) _hide_page_text_and_imagesr rectrotationZrotation_matrixfitzZMatrixZ get_pixmapparentitems update_stream) r rrrZ stream_dict clip_bboxrpixdocxrefstreamr r rclip_page_to_pixmaps   z#ImagesExtractor.clip_page_to_pixmapclip_image_res_ratiocCs|j|||d}|||S)aClip page pixmap (without text) according to ``bbox`` and convert to source image. Args: bbox (fitz.Rect, optional): Target area to clip. Defaults to None, i.e. entire page. rm_image (bool): remove images or not. clip_image_res_ratio (float, optional): Resolution ratio of clipped bitmap. Defaults to 3.0. Returns: list: A list of image raw dict. )rrr)r" _to_raw_dict)r rrr#rr r rclip_page_to_dictKs z!ImagesExtractor.clip_page_to_dictcCsp|durdSz|j|j|j|jf\}}}}Wn ty YdSwt||}tt|dd}t |dS)zExtract rotation angle (0, 90, 180, 270) from image transform matrix. Handles 90-degree multiples only. Based on PyMuPDF Matrix format: fitz.Matrix(90) -> (0, 1, -1, 0, ...), Matrix(180) -> (-1, 0, 0, -1, ...). NrZih) abcdAttributeErrormathatan2rounddegreesint)rr'r(r)r*Z angle_radZ angle_degr r r_get_image_rotationas    z#ImagesExtractor._get_image_rotationc Cs|jj}|jj}t}|jjddD]p}t|}d|d<|d}|j|}g}z |jj|dd}Wn ttfy=Ynw|jj } t |D];\} } | dkrQqF| | sWqFd} | t |kry|| } t| ttfryt | dkry| d}||} || || fqFqd d }||}g}|D]V}t |dkrt}|D] \} }}|| O}q||d |}n1|d\} }} |d d kr|| d |}n|||}||| }|pd| }|r||||d<||q|S)aExtract normal images with ``Page.get_images()``. Args: clip_image_res_ratio (float, optional): Resolution ratio of clipped bitmap. Defaults to 3.0. Returns: list: A list of extracted and recovered image raw dict. .. note:: ``Page.get_images()`` contains each image only once, which may less than the real count of images in a page. Tfullr) transformrcSs|d|dS)Nr) intersects)r'r(r r rsz0ImagesExtractor.extract_images..Fimage)r rrr get_imageslistZget_image_rects TypeErrorr+Zcropbox enumerateZget_arear8len isinstancetupler1appendgrouprRectr%_recover_pixmapr$ _rotate_image)r r#rrZicitemr ZrectsZrects_with_transformZunrotated_page_bboxirZimage_rotationentryZmatrix_tZfungroupsZimagesrEr_Zraw_dictrZtotal_rotationr r rextract_imagesssb              zImagesExtractor.extract_imagesmin_svg_gap_dxmin_svg_gap_dymin_wmin_hc sTddl}|jddd}||}|||j}||dd|j\} t||d} fd d | D} tt | | } d } | rt | D])\}\}}}}t |||||ddf||||f}| d ||qC| D]-\}}|\}}}}| |||f||fd d|D]\}}}}| |||f||fddqqo| d||d| S)aFind contour of potential vector graphics. Args: min_svg_gap_dx (float): Merge svg if the horizontal gap is less than this value. min_svg_gap_dy (float): Merge svg if the vertical gap is less than this value. min_w (float): Ignore contours if the bbox width is less than this value. min_h (float): Ignore contours if the bbox height is less than this value. Returns: list: A list of potential svg region: (external_bbox, inner_bboxes:list). rNT?)rr)Zmin_dxZmin_dycsg|] }t|qSr )r).0rbinaryrRrQr r sz7ImagesExtractor.detect_svg_contours..Fz sub-image-)rUrrr7)rrrUimg)cv2r"_pixmap_to_cv_imageZcvtColorZCOLOR_BGR2GRAY thresholdZTHRESH_BINARY_INVrr>zipr@rZimshowZ rectangleZwaitKey)r rOrPrQrRcvpixmapsrcgrayrMZexternal_bboxesZgrouped_inner_bboxesrLdebugrJx0y0x1y1ZarrrZ inner_bboxesZu0Zv0u1Zv1r rWrdetect_svg_contourss4 0    z#ImagesExtractor.detect_svg_contoursr<cCsN |j}|dus|tjtjfvrttj|}tjjt||j |j | dS)zStore Pixmap ``image`` to raw dict. Args: image (fitz.Pixmap): Pixmap to store. bbox (fitz.Rect): Boundary box the pixmap. Returns: dict: Raw dict of the pixmap. N)typerwidthheightr<) Z colorspacerZcsGRAYcsRGBPixmaprZIMAGEvaluerCrkrltobytes)r<rcsr r rr$ s zImagesExtractor._to_raw_dictr`rcCsddl}ddl}t|}|jdd\}}|d|d}}d} |||f|| } || d} || d} t|| || } t|| || }| d| d|7<| d|d|7<||| | |f}| d |\}}| S) zRotate image represented by image bytes. Args: pixmap (fitz.Pixmap): Image to rotate. rotation (int): Rotation angle. Return: image bytes. rNrrS)rr)rr7)rr)r7rz.png) r[numpyrr\shapeZgetRotationMatrix2Dabsr0Z warpAffineZimencoderp)r`rr_nprZhwrdrescalercossinWHZ rotated_imgrMZim_pngr r rrHMs   zImagesExtractor._rotate_imagerc sddD}|dd}fdd}j}i}|D]-}||} |r.|| n| df\} } |r:|| n| df\} } | sD| rN||| | ||<q!|S)zHide page text and images.cSsg|]\}}}}|qSr r )rVr nameZinvokerrr r rrYsz>ImagesExtractor._hide_page_text_and_images..cSsD|}d}dD]}|}||vrd}|||d}q||fS)NF)ZBTZTmZTdz2 TrTz 3 Tr)encodereplace)r!resfoundkbkr r r hide_textsz=ImagesExtractor._hide_page_text_and_images..hide_textcsX|}d}ddjddD}|D]}d|d}||vr'd}||d}q||fS) NFcSsg|]}|dqS)r )rVrIr r rrYszSImagesExtractor._hide_page_text_and_images..hide_images..Tr2/z Do)r=r~r)r!rrZ img_namesrrr r r hide_imagess z?ImagesExtractor._hide_page_text_and_images..hide_imagesF)Z get_xobjectsextendZ get_contentsrZ xref_streamr) r rrZ xref_listrrrsourcer rar!Z found_textZ found_imagesr rrrzs    z*ImagesExtractor._hide_page_text_and_imagesrrIcCs|d}|d}t||}|dkr>t||}|jr%t|d}d}|}|j|jkr8|j|jkr8t||}ntd|d|dvrMttj|}|S)aRestore pixmap with soft mask considered. References: * https://pymupdf.readthedocs.io/en/latest/document.html#Document.getPageImageList * https://pymupdf.readthedocs.io/en/latest/faq.html#how-to-handle-stencil-masks * https://github.com/pymupdf/PyMuPDF/issues/670 Args: doc (fitz.Document): pdf document. item (list): image instance of ``page.get_images()``. Returns: fitz.Pixmap: Recovered pixmap with soft mask considered. rr7NzCIgnore image due to inconsistent size of color and mask pixmaps: %sZCMYKr:) rrnalpharkrlloggingwarningupperrm)rrIxsrmasktempr r rrGs$   zImagesExtractor._recover_pixmapcCs0ddl}ddl}|}||||j|jS)znConvert fitz Pixmap to opencv image. Args: pixmap (fitz.Pixmap): PyMuPDF Pixmap. rN)r[rrrpZimdecodeZ frombufferZuint8Z IMREAD_COLOR)r`r_ruZimg_byter r rr\sz#ImagesExtractor._pixmap_to_cv_image)NFr)r)__name__ __module__ __qualname____doc__rZPagerrFboolfloatr"r% staticmethodr0r1rNrirnr$rHrZDocumentr>rGr\r r r rrsV  / r ;,,81r) rrr,rZcommon.CollectionrZ common.sharerZcommon.algorithmrrrrr r r rs