使用基于ORB的描述符进行图像聚类

问题描述

我一直在尝试使用多种方法进行图像聚类。我已经能够使用基于颜色的功能，并通过Keras和多种方式使用特征提取。我无法使用基于ORB的功能描述符成功完成此操作。

在下面提到的第二个代码段中，我尝试将其重塑为平坦的np.Array，但我无法。

如果我愿意 descriptors.append(np.array([orb_desc]))然后我得到 ValueError: Could not broadcast input array from shape (1027,32) into shape (1)

如果我做descriptors.append(np.array(orb_desc))，我会得到 TypeError: only size-1 arrays can be converted to Python scalars ValueError: setting an array element with a sequence.

在flat_descriptors = np.array(np.float32(descriptors).reshape(len(descriptors),-1)/255)行中都是

不确定如何格式化/重塑npArray以使其具有可接受的正确格式。

我的主要群集代码

def cluster(db_image_list,clusters=2):
    # load the image and convert it from BGR to RGB so that
    # we can display it with matplotlib
    images = [cv.resize(cv.imread(file),(224,224)) for file in db_image_list]
    images = [cv.cvtColor(image,cv.COLOR_BGR2RGB) for image in images]
    
    # predicted_images = get_color_images(images) # Only Colour based matrices do not get very satisfactory results
    # predicted_images = get_keras_prediction(images)
    # predicted_images = orb_feature_descriptor_detection(images)
    predicted_images = hog_histogram_images(images)

    clt = KMeans(n_clusters = clusters)
    clt.fit(predicted_images)
    kpredictions = clt.predict(predicted_images)

    for i in range(clusters):
        os.makedirs("output\cluster" + str(i))
    for i in range(len(db_image_list)):
        print("clustering copying",i)
        shutil.copy2(db_image_list[i],"output\cluster"+str(kpredictions[i]))

ORB功能

def orb_feature_descriptor_detection(images):
    # WIP. Can't figure out to reshape the descriptors properly
    descriptors = []
    for image in images:
        _,orb_desc = fd.orb_descriptor(image,False)
        if orb_desc is not None:
            descriptors.append(np.array([orb_desc]))

    flat_descriptors = np.array(np.float32(descriptors).reshape(len(descriptors),-1)/255)

    return flat_descriptors

基本ORB KP，描述符检索器

def orb_descriptor(img,bounding_rm=True):
    img_gray = cv.cvtColor(img,cv.COLOR_BGR2GRAY)
    orb = cv.ORB_create(2000)
    if bounding_rm:
        bounding = text_Boxes.detect_text_Box(img)
        bounding_mask = np.zeros(img_gray.shape,np.uint8)
        bounding_mask[bounding[1]:bounding[3],bounding[0]:bounding[2]] = 255
        bounding_mask = (255-bounding_mask)
        kp,des=orb.detectAndCompute(img_gray,mask = bounding_mask)
    else:
        kp,des = orb.detectAndCompute(img_gray,mask = None)

    return (kp,des)

解决方法

暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！

如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。

小编邮箱:dio#foxmail.com (将#修改为@）

numpy opencv orb python