Insights about OCR Vision Applications with CIJ Printers (Part 1 of 3)
One Monday morning I find this email in my mailbox sent by a customer:
"˜Hi Eldad, how is it going? Can you help me with this application: I have a CIJ printer installed on a high speed production line. Several lines of text are printed on each product. I need a vision system for inline inspection of the printed characters. I try using OCR tool, then compare the data read by the camera with a data string sent from the printer and finally generate a warning output signal when mismatch occurs. I find it hard to train the system with one font library to solve all products. I still have lots of false rejects, where the camera rejects good products. I'm on it for several days and still cannot make it work - can you advise?'
I often receive such applications from customers. Obviously the specifications are not complete, but this is not the issue I want to discuss in this article. I think that most of vision Engineers would admit that such a solution is not trivial to design. The point is that many times the requested specifications are not in line with the real needs of the customer. Then the Vision Engineer is struggling with the wrong problem. You can see below a typical CIJ print out and notice how much the characters are distorted. Therefore I have decided to write this article and share with others some insights I have learned recently while designing quite a few vision solutions for OCR with CIJ printers.
We will soon take a look at the challenges awaiting us with CIJ printers and vision, to have a better understanding about the difficulties involved when we try to design such a vision solution. Due to the technical nature of the printer together with the high speed production line, the print out of the text has very high variation in character size, character shape, orientation, scaling, position, etc. All these challenges are added to the other issues we might have with any application, like proper illumination, optical setup and any other integration related issues.
Now let's look at some images of text printed by a CIJ printer in high speed and analyze what we see. The images are taken by a Microscan Vision System which was installed recently for a major beverage factory. The text is printed on aluminum cans by CIJ printer in production rate of ~800 ppm.
Image example 1: From looking at this image below we can see the big difference between 2 similar digits which are printed in different locations on the can surface. One on the top line and the other one on the bottom line. Notice the pair of digits 2.
Image example 2: Notice this set of "˜1' digit occurrences in the image below. They are printed by the same printer head. The shapes are significantly different from each other.
Image example 3: Notice this pair of digit "˜5' occurrences, how different they are from each other as a result of displacement of just 2 critical dots in the bottom, which are ~15% from the complete character area. Also with a human eye I could read this as a "˜6' digit instead of "˜5'!
Such variations can be happening due to the high conveyor speed which is not strictly permanent, the bouncing of the can from side to side on the narrow conveyor and the non-flat surface of the can. Distance of print head from the object can be variable. All these factors are inevitable. We might face them on every production line.
Image example 4: The image below is a print out of character "˜M' on the left and the digit "˜1' on the right. Looking at it, if we split the "˜M' in the middle, it is composed of 2 mirrored digits of "˜1'. Isn't it easy to make the mistake and see there 2 consecutive digits of "˜1'? The mirrored "˜1' on the left can be considered as unidentified symbol.
Image example 5: These are 2 consecutive prints from the same printer. The second is printed one minute later. Notice the digit 1 on the left. One or two dots displacement brings great variation to the patterns of "˜1' digits which the system must identify.
Image example 6: In this image we can see a significant change in the size of the characters from left to right. This might happen due to the print location on the can, which is slightly changed here, together with the curved surface of the can - on the right side the printer head is closer to the object. On the left side the printer head is further away. Therefore the digits are smaller on the right side.
Now after we understand the nature of the printer output in this high speed production line and noticed the high variation between one print to the next print, we can realize why such applications are very challenging for any computer system which try to classify all occurrences correctly. The more differences there are within one character shapes, the harder it is to make a difference between one character to another character. Once we realize that insight, then we start to ask questions like:
1. "˜Do I really need the system to read every can and every character in 100%?'
2. "˜Do I really need the data to be read and compared to a match string from the printer?'
3. "˜How sensitive the production line is to false rejects? To overlooked defects?'
The answers to such questions will come in a later stage. Being able to answer those questions will lead usually to better specifications and better solution which is easy to design, install and change later. Having that in mind, we can soon begin with discussing the solution options. But first we must understand the nature of the defects which are relevant for the specific production line where Continuous Inkjet printer is used. These are dictating a wide range of possible vision solutions, which are not necessarily OCR applications. We are going to discuss all that and more, in my next articles which are going to be published shortly.
As a follow-up to this article, I will give examples of possible printing defect which are typical in CIJ printers and finally I will describe possible ideas of solutions. I will present highly advanced solutions which are introduced by Microscan in the latest software releases of AutoVISION.
Marked for Life: A Glance at Some Common Direct Part Marking Methods
Direct part marking is crucial for enabling the traceability of products and the parts comprising...
Do Your DPMs Make the Grade? Understanding the 2D Barcode Verification Parameters in ISO 15415
Direct part marks (DPMs) and other two-dimensional codes can vary widely in their readability. Fr...
How X-Mode Brings Out the Best in Barcodes
Barcodes have a unique challenge. Whether printed, etched, engraved or stamped onto their substra...