A key function of data capture systems often taken for granted is locating the desired objects on a form before they can be processed. These particular objects, known as "Regions of Interest" (ROI), contain the critical information customers need to use their application.
In the postal industry, ROI location is a necessary step in mail-forwarding and sorting applications pertaining to parcels, bundles, flats, and letters. Typical ROIs for this industry include destination address blocks, barcodes, identification codes, return addresses, indicia, etc., all of which may have independent ROI values or serve as landmarks that enable easier search for a target ROI.
The task of locating ROIs can be challenging. Take for example the front and back cover of a typical magazine: one large piece of paper featuring headlines, photos, advertisements, and logos in a variety of colors and sizes. Finding the small address block among these images must be approached in a very precise way. At the same time, the process must have the flexibility to accommodate different applications.
Following are some key considerations when performing ROI location tasks.
ROIs can have different meanings at various stages of the recognition process. To illustrate this concept, let's use location of a destination address block as an example. In this case ROI location includes three stages. First, the image of a mail piece should be separated from the background. Here, an ROI is the image of the mail piece, which is found within a larger image that also captures a piece of a conveyer belt used to transport the mail piece. The software separates the ROI, cleaning up the image for processing.
At the second stage - identifying the format- ROI represents the required information found in either a "simple" address block or a label containing an address block. For labels, which may contain additional information such as return address, phone number, account number, or barcode, a third stage is required.
Finally, the destination address block is located. While processing labels, a new ROI location task is assigned to extract the destination address data only. Sophisticated ROI algorithms filter out irrelevant objects at this stage.
One of the challenges in ROI location is associated with grayscale versus binary image identification. Grayscale images are the optimal format for conducting ROI location, from a quality perspective. Grayscale images contain much more information compared to corresponding binary images. They are frequently used in highly sensitive operations, such as the separation of handwritten text from their backgrounds.
However, grayscale imaging is a time-consuming task - even for modern computers. Therefore, binary imaging is used more frequently. In applications where the quality of images is often imperfect, such as parcel sorting, close interaction between grayscale and binary image representation may be required.
Another issue common to the ROI location process is orientation of both an image and an ROI within the image. In applications such as parcel sorting, either orientation may be arbitrary. Therefore, location algorithms should be multi-directional or have special algorithms that can determine the orientation, transform the image, and standardize it.
Modern technologies can successfully locate ROI on letters, flats, and parcels, even in difficult situations. The most advanced algorithms rely on two main approaches of the ROI location software:
First are heuristic algorithms, where a set of rules is created to describe the possible layouts of various elements of a mail piece. During the location process, an algorithm tries to classify an input layout by matching it against a set of fixed layout descriptions.
Another powerful method is based on neural net technologies, which are used to recognize relationships between items on a mail piece. The neural network is trained on a large image set of mail pieces. The accumulated knowledge is used to locate ROIs during the mail handling process.
Neural net technology offers the greatest potential for growth and improvement as we continue to enhance ROI location. To achieve the highest precision at the mail handling stage, we must consider the critical features and hidden drivers in this complex task. Therefore, ROI software must be flexible enough to locate different types of objects and adapt to multiple image types and layouts.
Dr. Tatyana Vazulina has worked in product management and marketing at Parascript, LLC., an address recognition and interpretation company, for over a decade. She can be reached at Tatyana.vazulina@parascript.com or (303)-381-3106.
In the postal industry, ROI location is a necessary step in mail-forwarding and sorting applications pertaining to parcels, bundles, flats, and letters. Typical ROIs for this industry include destination address blocks, barcodes, identification codes, return addresses, indicia, etc., all of which may have independent ROI values or serve as landmarks that enable easier search for a target ROI.
The task of locating ROIs can be challenging. Take for example the front and back cover of a typical magazine: one large piece of paper featuring headlines, photos, advertisements, and logos in a variety of colors and sizes. Finding the small address block among these images must be approached in a very precise way. At the same time, the process must have the flexibility to accommodate different applications.
Following are some key considerations when performing ROI location tasks.
ROIs can have different meanings at various stages of the recognition process. To illustrate this concept, let's use location of a destination address block as an example. In this case ROI location includes three stages. First, the image of a mail piece should be separated from the background. Here, an ROI is the image of the mail piece, which is found within a larger image that also captures a piece of a conveyer belt used to transport the mail piece. The software separates the ROI, cleaning up the image for processing.
At the second stage - identifying the format- ROI represents the required information found in either a "simple" address block or a label containing an address block. For labels, which may contain additional information such as return address, phone number, account number, or barcode, a third stage is required.
Finally, the destination address block is located. While processing labels, a new ROI location task is assigned to extract the destination address data only. Sophisticated ROI algorithms filter out irrelevant objects at this stage.
One of the challenges in ROI location is associated with grayscale versus binary image identification. Grayscale images are the optimal format for conducting ROI location, from a quality perspective. Grayscale images contain much more information compared to corresponding binary images. They are frequently used in highly sensitive operations, such as the separation of handwritten text from their backgrounds.
However, grayscale imaging is a time-consuming task - even for modern computers. Therefore, binary imaging is used more frequently. In applications where the quality of images is often imperfect, such as parcel sorting, close interaction between grayscale and binary image representation may be required.
Another issue common to the ROI location process is orientation of both an image and an ROI within the image. In applications such as parcel sorting, either orientation may be arbitrary. Therefore, location algorithms should be multi-directional or have special algorithms that can determine the orientation, transform the image, and standardize it.
Modern technologies can successfully locate ROI on letters, flats, and parcels, even in difficult situations. The most advanced algorithms rely on two main approaches of the ROI location software:
First are heuristic algorithms, where a set of rules is created to describe the possible layouts of various elements of a mail piece. During the location process, an algorithm tries to classify an input layout by matching it against a set of fixed layout descriptions.
Another powerful method is based on neural net technologies, which are used to recognize relationships between items on a mail piece. The neural network is trained on a large image set of mail pieces. The accumulated knowledge is used to locate ROIs during the mail handling process.
Neural net technology offers the greatest potential for growth and improvement as we continue to enhance ROI location. To achieve the highest precision at the mail handling stage, we must consider the critical features and hidden drivers in this complex task. Therefore, ROI software must be flexible enough to locate different types of objects and adapt to multiple image types and layouts.
Dr. Tatyana Vazulina has worked in product management and marketing at Parascript, LLC., an address recognition and interpretation company, for over a decade. She can be reached at Tatyana.vazulina@parascript.com or (303)-381-3106.
