Systems, methods and articles for reading highly blurred machine-readable symbols

Patent metadata

Field	Value
Publication number	US-2016125218-A1
Application number	US-201414528697-A
Country	US
Kind code	A1
Filing date	Oct 30, 2014
Priority date	Oct 30, 2014
Publication date	May 5, 2016
Grant date	—

Abstract

Systems and methods for robust recognition of machine-readable symbols from highly blurred or distorted images. An image signal representation of a machine-readable symbol element is transformed into a different space using one or more transform operations, which moves an n-dimensional vector of measured light intensities into another n-dimensional space. The types of transform operations may include blur robust orthonormal bases, such as the Discrete Sine Transform, the Discrete Cosine Transform, the Chebyshev Transform, and the Lagrange Transform. A trained classifier (e.g., an artificial intelligence machine learning algorithm) may be used to classify the transformed signal in the transformed space. The types of trainable classifiers that may be used include random forest classifiers, Mahalanobis classifiers, support vector machines, and classification or regression trees.

First claim

1 . A method of operation for a processor-based device to identify a machine-readable symbol in an image, the processor-based device including at least one processor and at least one nontransitory processor-readable storage medium, the method comprising:
receiving, in the at least one nontransitory processor-readable storage medium, a plurality of training images, each training image corresponding to one of a plurality of machine-readable symbols; generating, by the at least one processor, a distortion model for the training images; generating, by the at least one processor, a plurality of distorted image signals based at least in part on the distortion model and the training images, each of the plurality of distorted image signals corresponding to one of the machine-readable symbols; transforming, by the at least one processor, the plurality of distorted image signals from a signal space into a first transform space; extracting, by the at least one processor, classification features from the transformed distorted image signals in the first transform space; and training, by the at least one processor, a first machine-readable symbol classifier based at least in part on the extracted classification features.

2 . The method of claim 1 , further comprising:
receiving a run-time image in the at least one nontransitory processor-readable storage medium; transforming, by the at least one processor, image signals of the received run-time image from a signal space into the first transform space; extracting, by the at least one processor, classification features from the transformed image signals in the first transform space; and classifying the run-time image using the trained first machine-readable symbol classifier using the extracted classification features.

3 . The method of claim 2 , further comprising:
determining, by the at least one processor, a location of the received run-time image that includes a machine-readable symbol; and determining, by the at least one processor, a distortion effect of an object adjacent the location of the received run-time image that includes machine-readable symbol, wherein classifying the run-time image comprises accounting for the determined distortion effect.

4 . The method of claim 3 wherein determining a distortion effect of an object adjacent the location comprises determining a distortion effect of at least one of: a machine-readable symbol element, a start pattern, or a stop pattern.

5 . The method of claim 1 wherein extracting classification features from the distorted image signals in the first transform space comprises extracting a set of spectral coefficients.

6 . The method of claim 1 wherein transforming the plurality of distorted image signals from a signal space into a first transform space comprises performing a discrete cosine transform (DCT) on the distorted image signals, and extracting classification features from the distorted image signals in the first transform space comprises extracting a number of DCT coefficients obtained by performing the DCT on the distorted image signals.

7 . The method of claim 1 wherein generating a distortion model for the training images comprises generating a distortion model that accounts for at least one of: optical blur, ink spread, quantization shift, luminance variation, or sensor noise.

8 . The method of claim 1 wherein transforming the plurality of distorted image signals from a signal space into a first transform space comprises performing a transform that is robust to a low pass centrosymmetric filter.

9 . The method of claim 8 wherein performing a transform that is robust to a low pass centrosymmetric filter comprises performing a Discrete Sine Transform, a Discrete Cosine Transform, a Chebyshev Transform, or a Lagrange Transform.

10 . The method of claim 1 wherein extracting classification features from the distorted image signals in the first transform space comprises extracting a number of transform coefficients obtained by performing the transform on the distorted image signals.

11 . The method of claim 1 wherein training a first machine-readable symbol classifier based at least in part on the extracted classification features comprises training at least one of a random forest classifier or a Mahalanobis classifier.

12 . The method of claim 1 wherein receiving a plurality of training images comprises receiving a plurality of training images, each training image corresponding to a one-dimensional machine-readable symbol or a two-dimensional machine-readable symbol.

13 . The method of claim 1 , further comprising:
determining, by the at least one processor, a quality measure for the first machine-readable symbol classifier; transforming, by the at least one processor, the plurality of distorted image signals from the signal space into a second transform space; extracting, by the at least one processor, classification features from the distorted image signals in the second transform space; training, by the at least one processor, a second machine-readable symbol classifier based at least in part on the extracted classification features; determining, by the at least one processor, a quality measure for the second machine-readable symbol classifier; and selecting one of the first machine-readable symbol classifier or the second machine-readable symbol classifier based at least in part on the determined quality measure.

14 . The method of claim 13 wherein determining a quality measure for the first and the second machine-readable symbol classifiers comprises determining a measure of blur invariance for the first and the second machine-readable symbol classifiers.

15 . A method for detecting a machine-readable symbol, the method comprising:
acquiring, by at least one processor, an electronic representation of a machine-readable symbol image in at least one nontransitory processor-readable storage medium; transforming, by the at least one processor, the electronic representation of the machine-readable symbol image from a signal space to a transform space; extracting, by the at least one processor, machine-readable symbol feature vectors from the transformed electronic representation of the machine-readable symbol image; and classifying, by the at least one processor, the machine-readable symbol image using a classifier trained in a supervised manner from a dataset of simulated degraded machine-readable symbol feature vectors with a known class.

16 . An image processor system, comprising:
at least one processor; at least one nontransitory processor-readable storage medium operatively coupled to the at least one processor and storing at least one of data or instructions which, when executed by the at least one processor, cause the at least one processor to:
receive a plurality of training images, each of the plurality of training images corresponds to one of a plurality of machine-readable symbols;
generate a distortion model for the training images;
generate a plurality of distorted image signals based at least in part on the distortion model and the training images, each of the plurality of distorted image signals corresponds to one of the machine-readable symbols;
transform the plurality of distorted image signals from a signal space into a first transform space;
extract classification features from the transformed distorted image signals in the first transform space; and
train a first machine-readable symbol classifier based at least in part on the extracted classification features.

17 . The image processor system of claim 16

Assignees

• Datalogic IP Tech Srl

Inventors

• Deppieri Francesco
• De Girolami Maurizio Aldo
• Lanza Alessandro
• Sgallari Fiorella

Classifications

• G06K7/1473 Primary

  error correction · CPC title

• G06V30/2247 Primary

  Characters composed of bars, e.g. CMC-7 · CPC title

• G06V10/431

  Frequency domain transformation; Autocorrelation · CPC title