Assigning a semantically meaningful label to a digital image

A digital imaging system comprising a processor, a memory coupled to the processor, and a computer-readable hardware storage device coupled to the processor, the storage device containing program code configured to be run by the processor via the memory to implement a method for assigning a semantically meaningful label to a digital image, the method comprising:

• deriving from the digital image a set of feature vectors that each comprises numeric representations of one or more characteristics of the digital image, where the set of feature vectors comprises a set of pixel-intensity vectors and a set of learned-feature vectors;
• submitting the feature vectors to a support-vector machine that has been pretrained to associate each of the feature vectors with a possible type of subject matter of the digital image;
• receiving from the support-vector machine a candidate image label for each of the feature vectors, where each of the candidate image labels indicates a respective type of subject matter of the digital image;
• selecting a single image label of the candidate image labels by mathematically combining the received candidate image labels,
• wherein the mathematically combining comprises solving a linear equation, and wherein the linear equation comprises regression coefficients selected as a function of a pretraining of the convolutional neural network and further as a function of a pretraining of the support-vector machine;
• and
• associating the single image label with the digital image, where the single image label is a semantically meaningful label from which can be determined that the digital image depicts a certain type of subject matter.

A computer program product, comprising a computer-readable hardware storage device having a computer-readable program code stored therein, the program code configured to be executed by an image-identification system comprising a processor, a memory coupled to the processor, and an interface mechanism coupled to the processor for receiving a scanned image from a computer-tomography medical-imaging device, where the computer-readable program code, when executed by the image-identification system, implements a method for assigning a semantically meaningful label to a digital image, the method comprising:

• deriving from the digital image a set of feature vectors that each comprises numeric representations of one or more characteristics of the digital image, where the set of feature vectors comprises a set of pixel-intensity vectors and a set of learned-feature vectors;
• submitting the feature vectors to a support-vector machine that has been pretrained to associate each of the feature vectors with a possible type of subject matter of the digital image;
• receiving from the support-vector machine a candidate image label for each of the feature vectors, where each of the candidate image labels indicates a respective type of subject matter of the digital image;
• selecting a single image label of the candidate image labels by mathematically combining the received candidate image labels,
• wherein the mathematically combining comprises solving a linear equation, and wherein the linear equation comprises regression coefficients selected as a function of a pretraining of the convolutional neural network and further as a function of a pretraining of the support-vector machine;
• and
• associating the single image label with the digital image, where the single image label is a semantically meaningful label from which can be determined that the digital image depicts a certain type of subject matter.