1. A computer-implement method comprising:
obtaining a classification response map for an input image using a baseline detector;
generating a first-order descriptor, a first-order descriptor comprising a set of binary values in which each binary value is obtained by assigning a binary value based upon comparison of a pair of classification response values selected from the classification response map; and
generating a higher-order co-occurrence descriptor by performing the steps comprising:
calculating closeness vectors for a set of pairs of classification response values used to form the first-order descriptors relative to a local scan window;
generating a histogram of the closeness vectors; and
generating the higher-order co-occurrence descriptor as a cross product of the histogram.
2. The computer-implement method of claim 1 wherein:
a closeness vector measuring an absolute difference of locations of a pair of classification response values.
3. The computer-implement method of claim 1 further comprising:
forming a multi-order contextual co-occurrence descriptor comprising the first-order descriptor and the higher-order co-occurrence descriptor.
4. The computer-implement method of claim 3 further comprising:
evolving a detector trained using a descriptor comprising the multi-order contextual co-occurrence descriptor by iteratively adding an updated multi-order contextual co-occurrence descriptor to the descriptor at each iteration until a stop condition is reached.
5. The computer-implement method of claim 4 wherein a stop condition comprises at least one of:
convergence of the detector between iterations; and
a number of iterations have been reached.
6. The computer-implement method of claim 1 further comprising:
selecting pairs of classification response values to form the first-order descriptor is based upon a Gaussian distribution around a local scan window of interest.
7. The computer-implement method of claim 1 wherein the step of obtaining a classification response map for an input image using a baseline detector comprising:
using logistic regression on response values from the baseline detector to obtain the response map.
8. A system comprising:
one or more processors; and
a non-transitory computer-readable medium or media comprising one or more sequences of instructions which, when executed by the one or more processors, causes steps to be performed comprising:
obtaining a classification response map for an input image using a baseline detector;
generating a first-order descriptor, a first-order descriptor comprising a set of binary values in which each binary value is obtained by assigning a binary value based upon comparison of a pair of classification response values selected from the classification response map; and
generating a higher-order co-occurrence descriptor by performing the steps comprising:
calculating closeness vectors for a set of pairs of classification response values used to form the first-order descriptors relative to a local scan window;
generating a histogram of the closeness vectors; and
generating the higher-order co-occurrence descriptor as a cross product of the histogram.
9. The system of claim 8 wherein the one or more sequences of instructions further comprises:
a closeness vector measuring an absolute difference of locations of a pair of classification response values.
10. The system of claim 8 wherein the one or more sequences of instructions further comprises:
forming a multi-order contextual co-occurrence descriptor comprising the first-order descriptor and the higher-order co-occurrence descriptor.
11. The system of claim 10 wherein the one or more sequences of instructions further comprises:
evolving a detector trained using a descriptor comprising the multi-order contextual co-occurrence descriptor by iteratively adding an updated multi-order contextual co-occurrence descriptor to the descriptor at each iteration until a stop condition is reached.
12. The system of claim 11 wherein a stop condition comprises at least one of:
convergence of the detector between iterations; and
a number of iterations have been reached.
13. The system of claim 8 wherein the one or more sequences of instructions further comprises:
selecting pairs of classification response values to form the first-order descriptor is based upon a Gaussian distribution around a local scan window of interest.
14. The system of claim 8 wherein the step of obtaining a classification response map for an input image using a baseline detector comprising:
using logistic regression on response values from the baseline detector to obtain the response map.
15. The system of claim 8 wherein the step of obtaining a classification response map for an input image using a baseline detector comprising:
using a pre-trained model to obtain the detector response map that captures intra context.
16. The system of claim 15 wherein the pre-trained model is a deformable parts model and a plurality of detector response maps are generated.
17. A processor-based system comprising:
one or more processors; and
a non-transitory computer-readable medium or media comprising one or more sequences of instructions which, when executed by the one or more processors, causes steps to be performed comprising:
a computing a set of one or more image features for each local window of a set of local windows from the image:
b applying one or more pre-trained classifiers to the sets of one or more image features to obtain classifier response values;
c generating 0th-order context features using at least some of the classifier response values, a 0th-order context feature including classifier response values from a defined neighborhood;
d generating 1st-order context features using the 0th-order context features;
e generating higher-order context features using the 1st-order context features;
f forming multi-order co-occurrence (MOCO) features, wherein each MOCO corresponds to a local window and comprises two or more of the 0th-order context feature for that local window, 1st-order context feature for that local window, and higher-order context feature for that local window;
g forming contextual feature sets, wherein each contextual feature set corresponds to a local window and comprises a set of one or more image features for that local window and the MOCO for that local window; and
h applying a trained classifier to the contextual feature sets to obtain contextual classifier response values.
18. The processor-based system for classifying an object in an image of claim 17 further comprising:
responsive to a stop condition not being reached:
using the contextual classifier response values as the classifier response values in step b and repeating step c through f;
g\u2032 forming contextual feature sets, wherein each contextual feature set corresponds to a local window and comprises a set of one or more image features for that local window, the MOCO or MOCOs for that window from a prior iteration or iterations, and the MOCO for that local window for the current iteration; and
h applying a trained classifier corresponding to the current iteration to the contextual feature sets to obtain contextual classifier response values.
19. The processor-based system for classifying an object in an image of claim 17 wherein the neighborhood spans scan and location.
20. The processor-based system for classifying an object in an image of claim 17 further comprising:
responsive to the stop condition being reached, comparing the set of contextual classifier response values to one or more thresholds to decide whether the object was detected in local windows.
The claims below are in addition to those above.
All refrences to claim(s) which appear below refer to the numbering after this setence.
What is claimed is:
1. An image forming apparatus, wherein each interval of plurality of recording elements provided at a recording head in an intermittent relative feed direction is set such that irregularities in density, generated in an image, due to a feed amount error between a recorded material and the recording head in the intermittent relative feed direction, is generated in an area with a spatial frequency of 1 lpmm or more.
2. An image forming apparatus, wherein:
each interval of plurality of pixels in an intermittent relative feed direction, recorded substantially simultaneously by a plurality of recording elements which are provided at a recording head, is a natural number multiplied by d, wherein a minimum interval of the pixel in the intermittent relative feed direction is d;
a feed amount of the intermittent relative feed direction between a recorded material and the recording head is set to be (n2dk), wherein a number of the recording elements is n, k is a natural number of 2 or more; and
assuming that a pixel recorded by i th recording element is recorded on a position of the recorded material, which position is apart from a pixel recorded by a first recording element by a distance Li in the intermittent relative feed direction, each interval of plurality of the recording elements in the intermittent relative feed direction is set such that a pixel is not recorded on a position of the recorded material, which position is apart from the pixel recorded by the first recording element by a distance Lij(n2dk) in the intermittent relative feed direction, wherein j is a natural number.
3. The image forming apparatus according to claim 1, wherein a plurality of recording element rows, in which each of recording element rows the recording elements are arranged, are provided in a direction substantially perpendicular to the intermittent relative feed direction.
4. The image forming apparatus according to claim 2, wherein a plurality of recording element rows, in which each of recording element rows the recording elements are arranged, are provided in a direction substantially perpendicular to the intermittent relative feed direction.
5. The image forming apparatus according to claim 3, wherein the recording element rows are provided in at least two arrangement patterns.
6. The image forming apparatus according to claim 4, wherein the recording element rows are provided in at least two arrangement patterns.
7. The image forming apparatus according to claim 2, wherein the first recording element is an element positioned at one end portion of a recording element row in which the plurality of recording elements are arranged.
8. An image forming apparatus which forms an image by repeating for a plurality of times a relative movement of a recorded material and a recording portion for recording an image in a subscanning direction at a predetermined amount and a relative movement of the recorded material and the recording portion in the main scanning direction, the image forming apparatus comprising:
a plurality of recording elements which are provided at the recording portion and which are arranged in at least one row in the subscanning direction;
wherein the plurality of recording elements are arranged with non-uniform interval thereof in the subscanning direction.
9. The image forming apparatus according to claim 8, wherein distances, each of which is a distance between the recording elements which are adjacent in the subscanning direction, include at least three types of distances (ad), (bd) and (cd), wherein a, b, c are natural numbers which are different from each other, and a minimum value of the distance between adjacent recording elements is d.
10. The image forming apparatus according to claim 8, wherein the plurality of recording elements are arranged in a plurality of recording element rows arranged in the main scanning direction.
11. The image forming apparatus according to claim 10, wherein the plurality of recording element rows are provided such that arrangement patterns of at least two recording element rows among the plurality of recording element rows are different.
12. An image forming method for forming an image on a recorded material by a comb-teeth type printing by using an image forming apparatus, in which each interval of plurality of recording elements provided at a recording head in an intermittent relative feed direction is set such that irregularities in density, generated in an image, due to a feed amount error between the recorded material and the recording head in the intermittent relative feed direction, is generated in an area with a spatial frequency of 1 lpmm or more.