some time, the order of return text is not left to right, top to bottom

[gbrault]

When parsing a pdf text I would like to have it retruned left to right, top to bottom: it's not the case. I have created a function to do that an will push a pull request!

[softskills244]

When parsing a pdf text I would like to have it retruned left to right, top to bottom: it's not the case. I have created a function to do that an will push a pull request!

Did you push the request? I can't find it...I have the same issue

[gbrault]

yes I did: I will do it again

[gbrault]

Or best just put the code of the function here `def lrtd_parse_page(document,callback,context): """Parse page and yield text token left to right and top down (lrtd)

:param document: open stream to the PDF file to be worked on
:param callback: a function to callback each tim a page is processed accept 3 parameters 
     p the page, starts at 0 
     selts the lrtb token list and context.
     selts is a list of objects {"x1":x1,"y1":y1,"x0":x0,"y0":y0,"txt":text}
     (x0,y0),(x1,y1) are the coordinates of the box surrounding the object
     if callback return False, lrtd_parse_page quits
:param context: a caller context object for storing some data and passed to callback
"""
rsrcmgr = PDFResourceManager()
laparams = LAParams()
device = PDFPageAggregator(rsrcmgr, laparams=laparams)
interpreter = PDFPageInterpreter(rsrcmgr, device)
p=0
for page in PDFPage.get_pages(document):
        interpreter.process_page(page)
        layout = device.get_result()
        elts=[]
        m = 0
        mindeltaheight = 1000
        """
        breakdown lines into string (w/o \n) calculate new coordinates
        """
        for element in layout:
            if isinstance(element, LTTextBoxHorizontal):
                x0 = element.x0
                y0 = element.y0
                x1 = element.x1
                y1 = element.y1
                lines = element.get_text().splitlines()
                lenlines = len(lines)
                j = 0
                deltaheight = (y1-y0)/lenlines
                if mindeltaheight > deltaheight:
                    mindeltaheight = deltaheight
                for line in lines:
                    x0j = x0
                    y0j = y1 - (1+j)*deltaheight
                    x1j = x1
                    y1j = y1 - j*deltaheight
                    j += 1
                    elts.append({"x1":x1j,"y1":y1j,"x0":x0j,"y0":y0j,"txt":line})
                if m < element.y1:
                    m = element.y1
        n = len(elts)
        """
        tune strings coordinate to get them aligned in the same "line" if not too far apart 
        (less than 1/2 the min line height of the page)
        """
        for i in range(1,n):
            for j in range(i+1,n):
                if abs(elts[i-1]["y0"]-elts[j-1]["y0"])<(mindeltaheight/2):
                    elts[j-1]["y0"] = elts[i-1]["y0"]
                if abs(elts[i-1]["y1"]-elts[j-1]["y1"])<(mindeltaheight/2):
                    elts[j-1]["y1"] = elts[i-1]["y1"]
                
        selts = sorted(elts, key=lambda item: (round((2*m-item['y0']-item['y1'])/2,0),round(item['x0'],0)))   
        
        # for elt in selts:
        #    print(f"({p})({elt['x0']:0.0f},{elt['y0']:0.0f},{elt['x1']:0.0f},{elt['y1']:0.0f}){elt['txt']}")
        if not callback(p,selts,context):
            return
            
        p +=1`

I put this function in high_level.py. It does the following thinks:

• for the text block with multiple lines it breaks them into more single lines making interpolation
• It cluster the (x,y,x1,y1,text) tuples to have them belong to 'lines'
• it sort them left to right, top to bottom

Not 100% bullet proof but got good results over many sampled pdf files