Image quality assessment using adaptive non-overlapping mean estimation
Methods, apparatus, systems and articles of manufacture (e.g., physical storage media) to assess image quality using adaptive non-overlapping mean estimation are disclosed. Example apparatus disclosed herein include a machine learning system to be trained to classify image quality. Disclosed example apparatus also include a feature extractor to apply a blur filter to a first image to determine a blurred image, determine a blur feature value for the first image, the blur feature value to represent an amount the first image differs from the blurred image, and apply a vector of feature values associated with the first image to the machine learning system, the vector of feature values including the blur feature value. Disclosed example apparatus further include an image classifier to classify image quality associated with the first image based on an output of the machine learning system responsive to the vector of feature values associated with the first image.
1. An apparatus to perform image quality assessment, the apparatus comprising:
at least one memory including instructions; and
processor circuitry to execute the instructions to at least:
implement a machine learning system trained to classify image quality;
apply a blur filter to a first image to determine a blurred image;
determine a blur feature value for the first image, the blur feature value to represent an amount the first image differs from the blurred image;
segment the first image into blocks of pixels;
remove respective center pixels of the blocks of pixels to determine modified blocks of pixels;
determine a contrast feature value for the first image based on a mean of the modified blocks of pixels;
apply a vector of feature values associated with the first image to the machine learning system, the vector of feature values including the blur feature value and the contrast feature value; and
classify image quality associated with the first image based on an output of the machine learning system responsive to the vector of feature values associated with the first image.
2. The apparatus of claim 1 , wherein the blur filter includes a plurality of horizontal filter values and a plurality of vertical filter values, and the processor circuitry is to apply the blur filter along horizontal and vertical directions of the first image.
3. The apparatus of claim 2 , wherein a number of the horizontal filter values equals a number of the vertical filter values.
4. The apparatus of claim 1 , wherein the machine learning system is a back propagation neural network.
5. The apparatus of claim 1 , wherein the processor circuitry is to:
classify the first image as acceptable when the output of the machine learning system responsive to the vector of feature values satisfies a threshold; and
classify the first image as unacceptable when the output of the machine learning system responsive to the vector of feature values does not satisfy the threshold.
6. A tangible computer readable storage medium comprising computer readable instructions that, when executed, cause a processor to at least:
apply a blur filter to a first image to determine a blurred image;
determine a blur feature value for the first image, the blur feature value to represent an amount the first image differs from the blurred image;
segment the first image into blocks of pixels;
remove respective center pixels of the blocks of pixels to determine modified blocks of pixels;
determine a contrast feature value for the first image based on a mean of the modified blocks of pixels;
apply a vector of feature values associated with the first image to a machine learning system, the vector of feature values including the blur feature value and the contrast feature value, the machine learning system trained to classify image quality; and
classify image quality associated with the first image based on an output of the machine learning system responsive to the vector of feature values associated with the first image.
7. The tangible computer readable storage medium of claim 6 , wherein the blur filter includes a plurality of horizontal filter values and a plurality of vertical filter values, and the instructions, when executed, cause the processor to apply the blur filter along horizontal and vertical directions of the first image.
8. The tangible computer readable storage medium of claim 7 , wherein a number of the horizontal filter values equals a number of the vertical filter values.
9. The tangible computer readable storage medium of claim 6 , wherein the machine learning system is a back propagation neural network.
10. The tangible computer readable storage medium of claim 6 , wherein the instructions, when executed, cause the processor to:
classify the first image as acceptable when the output of the machine learning system responsive to the vector of feature values satisfies a threshold; and
classify the first image as unacceptable when the output of the machine learning system responsive to the vector of feature values does not satisfy the threshold.
11. An image quality assessment method comprising:
applying, by executing an instruction with a processor, a blur filter to a first image to determine a blurred image;
determining, by executing an instruction with the processor, a blur feature value for the first image, the blur feature value to represent an amount the first image differs from the blurred image;
segmenting, by executing an instruction with the processor, the first image into blocks of pixels;
removing, by executing an instruction with the processor, respective center pixels of the blocks of pixels to determine modified blocks of pixels;
determining, by executing an instruction with the processor, a contrast feature value for the first image based on a mean of the modified blocks of pixels;
applying, by executing an instruction with the processor, a vector of feature values associated with the first image to a machine learning system, the vector of feature values including the blur feature value and the contrast feature value, the machine learning system trained to classify image quality; and
classifying, by executing an instruction with the processor, image quality associated with the first image based on an output of the machine learning system responsive to the vector of feature values associated with the first image.
12. The method of claim 11 , wherein the blur filter includes a plurality of horizontal filter values and a plurality of vertical filter values, and further including applying the blur filter along horizontal and vertical directions of the first image.
13. The method of claim 12 , wherein a number of the horizontal filter values equals a number of the vertical filter values.
14. The method of claim 12 , further including:
classifying the first image as acceptable when the output of the machine learning system responsive to the vector of feature values satisfies a threshold; and
classifying the first image as unacceptable when the output of the machine learning system responsive to the vector of feature values does not satisfy the threshold.
15. The method of claim 12 , wherein the machine learning system is a back propagation neural network.