Image correction method, image correction device, and program
An image correction device includes a scene belonging rate computation unit that computes, from the feature value of an input image, a plurality of scene belonging rates each of which prescribes a probability with which the input image belongs to each category scene; an unknown scene belonging rate specification unit that specifies an unknown scene belonging rate prescribing a probability with which the input image belongs to an unknown scene; a by-scene correction parameter memory unit that stores by-scene correction parameters; a correction parameter combination unit that computes a combined correction parameter that is the weighted average of the correction parameters using the plurality of scene belonging rates and the unknown scene belonging rate; and an image correction unit that performs image correction processing for the input image using the combined correction parameter.
1. An image correction method comprising:
computing, from a feature value of an input image, a plurality of scene belonging rates each of which prescribes a probability with which the input image belongs to each category scene;
computing a combined correction parameter that is a weighted average of by-scene correction parameters using the plurality of scene belonging rates and an unknown scene belonging rate that prescribes a probability with which the input image belongs to an unknown scene; and
performing image correction processing for the input image using the combined correction parameter;
wherein said computing a combined correction parameter computation includes computing the combined correction parameter by further using a by-scene erroneous class cation weight value.
2. The image correction method as defined by claim 1 , wherein the correction parameter is an amount of color saturation enhancement.
3. The image correction method as defined by claim 1 , wherein the weight value takes a larger value as erroneous recognition has a greater effect on an image quality.
4. An image correction device comprising:
a scene belonging rate computation unit that computes, from a feature value of an input image, a plurality of scene belonging rates each of which prescribes a probability with which the input image belongs to each category scene;
an unknown scene belonging rate specification unit that specifies an unknown scene belonging rate prescribing a probability with which the input image belongs to an unknown scene;
a by-scene correction parameter memory unit that stores by-scene correction parameters;
a correction parameter combination unit that computes a combined correction parameter that is a weighted average of the correction parameters using the plurality of scene belonging rates and the unknown scene belonging rate;
an image correction unit that performs image correction processing for the input image using the combined correction parameter; and
a by-scene erroneous classification weight memory that stores a by-scene erroneous classification weight value, wherein said correction parameter combination unit computes the combined correction parameter by further using the weight value.
5. A non-transitory computer-readable medium storing a program that causes a computer to function as:
a scene belonging rate computation unit that computes, from a feature value of an input image, a plurality of scene belonging rates each of which prescribes a probability with which the input image belongs to each category scene; and
a correction parameter combination unit that computes a combined correction parameter that is a weighted average of by-scene correction parameters using the plurality of scene belonging rates and an unknown scene belonging rate that prescribes a probability with which the input image belongs to an unknown scene;
wherein said program further causes the computer to function in such away that said correction parameter combination unit computes the combined correction parameter by further using a by-scene erroneous classification weight value.