Recognition process of an object in a query image

A computer implemented recognition process of an object in a query image, comprising:

• performing a training step that comprises:
• providing a set of training images, each training image being defined by a plurality of pixels and comprising an object tag;
• determining for each training image of said set a plurality of first descriptors, each first descriptor being a vector that represents pixel properties in a corresponding subregion of the associated training image;
• selecting among the first descriptors a group of exemplar descriptors describing said set of training images, wherein selecting the exemplar descriptors comprises determining the first descriptors having a number of repetitions in the set of training images higher than a certain value;
• storing said group of exemplar descriptor in an exemplars library by associating to each exemplar descriptor:
• a respective pixel property vector;
• a geometric reference including a number of p parameters defining an enclosing region around the exemplar descriptor that is capable of enclosing the object tag in the training image associated with the exemplar descriptor;
• and a logo-code that comprises the object tag in the training image associated with the exemplar descriptor;
• and
• the method further comprises performing a query step, the query step comprising:
• receiving the query image and defining a plurality of second descriptors, the second descriptors being vectors describing pixel properties in corresponding subregions of the query image;
• determining visual similarity coefficients based on a comparison between each one of said second descriptors of the query image and each one of said exemplar descriptors;
• determining whether second descriptors of the query image match corresponding exemplar descriptors based on the determined visual similarity coefficients;
• recognizing the object in the query image based on the determined matches;
• the method further comprises performing a testing step, the testing step comprising:
• selecting in the exemplars library the geometric reference associated with the matched exemplar descriptors;
• generating a p-dimensional Vector Space, wherein each vector of the Vector Space represents one of the p-parameters of the geometric reference;
• determining a list of positive matched exemplar descriptors resulting from a geometric voting procedure for each second descriptor in said Vector Space, the geometric voting procedure identifying a region in the query image having a concentration of second descriptors high enough to consider the corresponding matched exemplar descriptors as positive exemplar descriptors;
• and
• comparing the object tag associated with at least one of the positive matched exemplar descriptors in said list with said object recognized during the query step in order to verify the object recognition.

A computer implemented recognition process of an object in a query image, comprising:

• performing a training step that comprises:
• providing a set of training images, each training image being defined by a plurality of pixels and comprising an object tag;
• determining for each training image of said set a plurality of first descriptors, each first descriptor being a vector that represents pixel properties in a corresponding subregion of the associated training image;
• selecting among the first descriptors a group of exemplar descriptors describing said set of training images, wherein selecting the exemplar descriptors comprises determining the first descriptors having a number of repetitions in the set of training images higher than a certain value;
• storing said group of exemplar descriptor in an exemplars library by associating to each exemplar descriptor:
• a respective pixel property vector;
• a geometric reference including a number of p parameters defining an enclosing region around the exemplar descriptor that is capable of enclosing the object tag in the training image associated with the exemplar descriptor;
• and a logo-code that comprises the object tag in the training image associated with the exemplar descriptor;
• and
• the method further comprises performing a query step, the query step comprising:
• receiving the query image and defining a plurality of second descriptors, the second descriptors being vectors describing pixel properties in corresponding subregions of the query image;
• determining visual similarity coefficients based on a comparison between each one of said second descriptors of the query image and each one of said exemplar descriptors;
• determining whether second descriptors of the query image match corresponding exemplar descriptors based on the determined visual similarity coefficients;
• recognizing the object in the query image based on the determined matches;
• and
• the method further comprises performing a testing step, the testing step comprising:
• selecting in the exemplars library the geometric reference associated with the matched exemplar descriptors;
• generating a p-dimensional Vector Space, wherein each vector of the Vector Space represents one of the p-parameters of the geometric reference;
• determining a list of positive matched exemplar descriptors resulting from a geometric voting procedure for each second descriptor in said Vector Space, the geometric voting procedure identifying a region in the query image having a concentration of second descriptors high enough to consider the corresponding matched exemplar descriptors as positive exemplar descriptors;
• determining a coverage index indicative of a percentage of the area of the object to be recognized in the query image which is covered by an area defined by the subregions of the positive matched exemplar descriptors, wherein:
• in response to determining that the coverage index is below a first threshold, identifying the positive matched exemplar descriptors as false positive matched exemplar descriptors;
• in response to determining that the coverage index is above a second threshold greater than the first threshold:
• confirming the positive matched exemplar descriptors;
• and
• verifying an object as being recognized if the object tag associated with at least one of the confirmed positive matched exemplar descriptors matches the object recognized during the query step;
• and
• in response to determining that the coverage index is between the first and second thresholds, identifying that the query image contains a candidate glimpse of the object.